Glossary
Molecular Informatics and Bio-Artificial Intelligence
Graph Neural Networks for Molecules
Terms related to geometric deep learning and equivariant neural networks applied to molecular graphs. Target: CTOs and R&D leaders in computational chemistry.
Message Passing Neural Network (MPNN)
A general framework for graph neural networks where node representations are iteratively updated by aggregating information from neighboring nodes via message and update functions.
Graph Isomorphism Network (GIN)
A theoretically maximally powerful GNN architecture under the Weisfeiler-Lehman test, designed to capture graph structure by learning injective aggregation functions over node neighborhoods.
SE(3) Equivariance
A property of a function or neural network ensuring that its output transforms consistently with any input rotation and translation in 3D Euclidean space, critical for modeling molecular geometries.
Graph Attention Network (GAT)
A GNN variant that employs self-attention mechanisms to assign learnable importance weights to different neighboring nodes during message aggregation.
Weisfeiler-Lehman Test
A classical graph isomorphism heuristic that iteratively refines node labels based on their neighbors, serving as the theoretical upper bound for the discriminative power of message-passing GNNs.
Equivariant Graph Neural Network (EGNN)
A computationally efficient architecture that achieves E(n) equivariance without requiring expensive higher-order tensor products, operating directly on scalar and vector features.
Molecular Fingerprint
A fixed-length bit vector encoding the presence or absence of specific substructural features within a molecule, used as a standardized input representation for machine learning models.
Neural Network Potential (NNP)
A machine-learned surrogate model that predicts the potential energy and atomic forces of a molecular system directly from atomic coordinates, bypassing the explicit solution of the Schrödinger equation.
Graph Convolutional Network (GCN)
A foundational GNN model that performs spectral or spatial convolution operations on graph-structured data, updating node features via a normalized sum of neighbor features.
SchNet
A pioneering continuous-filter convolutional neural network that models quantum interactions by using interatomic distances to generate filter kernels for message passing.
Equiformer
A transformer architecture that integrates SE(3)/E(3) equivariance using tensor products and equivariant attention, achieving state-of-the-art performance on 3D molecular property prediction.
Tensor Field Network
A locally equivariant neural network layer that builds upon point clouds by learning to map geometric point features to higher-order tensor fields using learned filters conditioned on relative positions.
Graph U-Net
An encoder-decoder architecture for graphs that performs graph pooling and unpooling operations, enabling the learning of hierarchical representations for tasks like graph classification.
Differentiable Pooling (DiffPool)
A learnable graph pooling method that hierarchically clusters nodes into a smaller graph using a soft assignment matrix, allowing end-to-end training for whole-graph prediction tasks.
SELFIES
A 100% robust molecular string representation where every syntactically valid string corresponds to a valid molecular graph, eliminating the invalid outputs common with SMILES-based generative models.
Graphormer
A standard transformer model adapted for graph data by encoding structural and spatial information, such as node centrality and shortest-path distances, into the attention mechanism.
Atomic Cluster Expansion (ACE)
A systematic and complete basis set expansion of atomic environments that yields highly efficient, body-ordered invariant features for constructing linear and neural network interatomic potentials.
SOAP Descriptor
A smooth, invariant representation of local atomic environments based on a Gaussian smeared atomic density expanded in radial and spherical harmonic basis functions.
Graph Contrastive Learning (GraphCL)
A self-supervised pre-training framework for GNNs that maximizes mutual information between differently augmented views of the same graph to learn robust, transferable representations.
GNNExplainer
A model-agnostic explainability tool that identifies the most relevant subgraph structure and node features contributing to a GNN's specific prediction.
MACE
A highly accurate equivariant message-passing interatomic potential that leverages many-body expansions via higher-order tensor products, achieving state-of-the-art efficiency and accuracy.
NequIP
An E(3)-equivariant neural network interatomic potential that uses tensor products of irreducible representations to achieve data-efficient and highly accurate force and energy predictions.
Graph Edit Distance (GED)
A graph similarity metric quantifying the minimum number of edit operations (insertions, deletions, substitutions of nodes/edges) required to transform one graph into another.
Open Catalyst Project (OC20)
A large-scale dataset and benchmark challenge for training machine learning models to predict the adsorption energy of molecules on catalyst surfaces for renewable energy applications.
Crystal Graph Convolutional Neural Network (CGCNN)
A GNN architecture that directly learns material properties from crystal structures by constructing a multigraph representing atomic connectivity across periodic boundary conditions.
Deep Graph Library (DGL)
A popular open-source Python library that provides a flexible and efficient framework for implementing and training graph neural networks on top of existing deep learning backends.
Jumping Knowledge Network
A GNN architecture that aggregates node representations from all intermediate layers to combine local and global neighborhood information, preventing over-smoothing in deep networks.
Equivariant Diffusion Model (EDM)
A generative model that learns to reverse a noising process on 3D atomic coordinates while maintaining E(3) or SE(3) equivariance, used for generating stable molecular conformers.
Torsional Diffusion
A diffusion-based generative model operating on the torsional angle space of molecules to efficiently generate physically realistic 3D conformers by only modifying rotatable bonds.
EquiBind
An SE(3)-equivariant deep learning model that performs fast, direct-shot prediction of a ligand's bound pose in a protein pocket without relying on iterative sampling or scoring.
Protein Structure Prediction
Terms related to AI models for predicting 3D protein folding, dynamics, and sequence-to-structure relationships. Target: CTOs and heads of structural biology.
AlphaFold2
A deep learning model developed by DeepMind that predicts protein 3D structure from amino acid sequence with atomic accuracy using a novel neural network architecture and evolutionary information.
RoseTTAFold
A three-track neural network architecture for protein structure prediction that simultaneously processes sequence, distance, and coordinate information, enabling high-accuracy modeling without requiring deep multiple sequence alignments.
Multiple Sequence Alignment (MSA)
A computational method that aligns three or more biological sequences to identify regions of evolutionary conservation and homology, serving as a critical input feature for modern protein structure prediction models.
Homology Modeling
A computational technique that predicts a protein's three-dimensional structure based on its sequence similarity to one or more experimentally determined template structures.
Protein Language Model
A transformer-based neural network trained on massive protein sequence databases that learns the underlying grammar of amino acid sequences to generate contextualized representations for structure and function prediction.
pLDDT (Predicted Local Distance Difference Test)
A per-residue confidence metric output by AlphaFold2 that estimates the local accuracy of the predicted structure on a scale from 0 to 100, with higher scores indicating higher reliability.
PAE (Predicted Aligned Error)
A pairwise confidence metric that estimates the expected positional error between any two residues in a predicted protein structure, used to assess domain packing and relative domain orientation accuracy.
TM-score
A scale-invariant metric for assessing the topological similarity between two protein structures, designed to be independent of protein size and to provide an intuitive measure of structural fold agreement.
RMSD (Root Mean Square Deviation)
A standard quantitative measure of the average distance between the backbone atoms of superimposed protein structures, used to evaluate the accuracy of predicted models against experimental references.
Ramachandran Plot
A two-dimensional scatter plot of the backbone dihedral angles phi and psi for each residue in a protein structure, used to validate the stereochemical quality and identify energetically disallowed conformations.
Equivariant Neural Network
A neural network architecture that guarantees its predictions transform predictably under 3D rotations and translations of the input coordinates, ensuring physically consistent protein structure representations.
IPA (Invariant Point Attention)
A core architectural component of AlphaFold2 that performs attention over 3D spatial relationships between residues while maintaining invariance to global rotation and translation of the protein structure.
Recycling
An iterative refinement mechanism in AlphaFold2 where the model's initial predicted structure is fed back as input for multiple passes, progressively improving the accuracy of the final 3D coordinates.
Denoising Diffusion Probabilistic Model (DDPM)
A class of generative models that learn to reverse a gradual noising process, recently applied to protein structure prediction to generate diverse conformational ensembles by iteratively denoising random atomic coordinates.
ProteinMPNN
A message-passing neural network designed for inverse protein folding that predicts amino acid sequences likely to fold into a given backbone structure, enabling robust sequence design for novel protein scaffolds.
Inverse Folding
The computational task of predicting an amino acid sequence that will stably fold into a specified three-dimensional protein backbone structure, the reverse problem of structure prediction.
CASP (Critical Assessment of Structure Prediction)
A biennial community-wide blind experiment that rigorously benchmarks the state-of-the-art in computational protein structure prediction methods against newly solved but unpublished experimental structures.
PDB (Protein Data Bank)
The single worldwide repository of experimentally determined three-dimensional structures of biological macromolecules, serving as the primary training and evaluation data source for structure prediction models.
Co-Evolutionary Analysis
A statistical method that identifies pairs of residues that have mutated in a correlated manner across evolution, providing spatial proximity constraints used to guide ab initio protein structure prediction.
Energy Minimization
A computational refinement procedure that adjusts atomic coordinates to find the nearest local minimum on a physics-based potential energy surface, relieving steric clashes and bond geometry violations in predicted structures.
Side-Chain Packing
The computational task of determining the optimal discrete rotameric state for each amino acid side chain on a fixed backbone scaffold to minimize steric overlap and maximize favorable interactions.
Conformational Ensemble
A collection of structurally distinct states representing the intrinsic dynamic flexibility of a protein, moving beyond a single static prediction to capture the functional motions relevant to binding and catalysis.
Intrinsically Disordered Region (IDR)
A segment of a protein that lacks a stable folded three-dimensional structure under physiological conditions, existing as a dynamic conformational ensemble and often mediating critical signaling interactions.
ESMFold
A large-scale protein language model developed by Meta AI that predicts atomic-resolution structures directly from single sequences without requiring multiple sequence alignments, enabling rapid metagenomic-scale structure prediction.
MolProbity
A widely used structure-validation software that analyzes all-atom contacts, hydrogen bonding, and backbone geometry to generate a clashscore and Ramachandran statistics for assessing the physical realism of protein models.
Oligomeric State
The number and arrangement of individual protein subunits that assemble to form a functional multi-chain complex, a critical prediction target for understanding biological assembly and quaternary structure.
DockQ
A continuous quality score for evaluating protein-protein docking predictions that combines interface RMSD, fraction of native contacts, and ligand RMSD into a single metric ranging from 0 to 1.
Folding Free Energy (ΔΔG)
The change in thermodynamic stability of a protein upon mutation, calculated as the difference in Gibbs free energy of folding between the mutant and wild-type sequences, used to predict variant effects.
Deep Mutational Scanning
A high-throughput experimental technique that measures the functional impact of thousands of protein sequence variants simultaneously, generating rich datasets for training and validating variant effect prediction models.
Fragment Assembly
A protein structure prediction methodology that builds models by sampling and assembling short peptide fragments from experimentally determined structures, historically foundational to the Rosetta software suite.
De Novo Drug Design
Terms related to generative chemistry models for creating novel molecular entities with optimized properties. Target: CTOs and heads of medicinal chemistry.
De Novo Molecular Generation
The computational design of novel chemical entities from scratch using generative models, without relying on existing compound libraries as starting templates.
Molecular Graph Generation
A generative modeling approach that constructs molecules atom-by-atom and bond-by-bond as graph structures, ensuring chemical validity through iterative node and edge addition.
SMILES-Based Generative Models
Deep learning architectures, typically recurrent neural networks or transformers, trained to generate syntactically valid molecular string representations for novel compound design.
Molecular VAE
A variational autoencoder that learns a continuous latent representation of molecular structures, enabling smooth interpolation and gradient-based optimization of chemical properties.
Molecular GAN
A generative adversarial network framework where a generator proposes novel molecular structures and a discriminator evaluates their resemblance to real drug-like molecules.
Reinforcement Learning for Molecular Design
A framework that treats molecular generation as a sequential decision process, where an agent is rewarded for producing structures with desired physicochemical and biological properties.
Multi-Objective Molecular Optimization
The simultaneous optimization of multiple, often conflicting, drug properties—such as potency, solubility, and synthetic accessibility—using Pareto frontier algorithms.
Chemical Space Exploration
The systematic navigation of the vast theoretical universe of synthesizable molecules to identify regions with high probability of containing viable drug candidates.
Scaffold Hopping
The computational identification of structurally novel core templates that retain the biological activity of a known compound while circumventing existing intellectual property.
Fragment-Based Generation
A molecular design strategy that computationally assembles novel ligands by linking or growing small, low-molecular-weight fragments with high binding efficiency.
Junction Tree Variational Autoencoder
A generative model that operates on a tree decomposition of molecular graphs, generating valid substructures before assembling them into complete molecules to ensure chemical validity.
Conditional Molecular Generation
The targeted generation of molecular structures with pre-specified property profiles, such as logP or binding affinity, by conditioning the generative model on numerical constraints.
Synthetic Accessibility Score
A quantitative metric, often derived from retrosynthetic complexity or fragment frequency, that estimates the ease with which a computationally designed molecule can be synthesized in the lab.
Quantitative Estimate of Drug-Likeness
A numerical score quantifying how closely a molecule's physicochemical properties align with those of known oral drugs, often based on distributions of molecular descriptors.
ADMET Property Prediction
The use of machine learning models to forecast a molecule's absorption, distribution, metabolism, excretion, and toxicity profiles early in the drug design process.
Molecular Fingerprint
A fixed-length bit or count vector encoding the presence or absence of specific substructures, used as a numerical input representation for predictive machine learning models.
Bayesian Optimization for Molecules
A sequential model-based optimization strategy that efficiently explores chemical space by balancing exploitation of high-scoring regions with exploration of uncertain ones.
Diversity-Promoting Loss
A regularization term added to generative model training that penalizes the production of similar molecules, ensuring the generated library covers a wide area of chemical space.
Tanimoto Similarity
A widely used metric for quantifying the structural similarity between two molecules based on the overlap of their molecular fingerprints, ranging from 0 to 1.
Chemical Validity Checker
A post-processing filter that verifies generated SMILES strings or graphs against basic chemical rules, such as correct valence and aromaticity, to discard invalid outputs.
Molecular Grammar
A formal set of production rules defining the syntax of valid chemical structures, used to constrain generative models to produce only chemically sensible molecules.
Monte Carlo Tree Search for Chemistry
A heuristic search algorithm that builds a search tree of molecular modifications, balancing random exploration with directed exploitation to optimize a chemical scoring function.
Active Learning Loop
An iterative design cycle where a predictive model identifies the most informative molecules to synthesize and assay next, rapidly converging on optimal candidates.
Inverse QSAR
The process of deriving novel molecular structures directly from a desired biological activity profile by inverting a quantitative structure-activity relationship model.
Lead Optimization
The late-stage drug discovery phase where a confirmed hit compound is systematically modified to improve its potency, selectivity, and pharmacokinetic profile before preclinical testing.
Focused Library Generation
The computational enumeration of a targeted set of virtual compounds designed around a specific scaffold or pharmacophore to explore structure-activity relationships efficiently.
Reaction-Based Generation
A generative approach that constructs molecules by applying known chemical reaction rules to available building blocks, ensuring the outputs are synthetically tractable by design.
Transfer Learning for Chemistry
A technique where a generative model pre-trained on a large generic molecular dataset is fine-tuned on a small set of active compounds to bias generation toward a specific target.
Design-Make-Test-Analyze Cycle
The iterative, closed-loop workflow in drug discovery where computational design, chemical synthesis, biological assay, and data analysis inform subsequent rounds of optimization.
Molecular Representation Learning
The use of self-supervised or contrastive learning to derive dense vector embeddings of molecules that capture meaningful chemical features for downstream predictive tasks.
Molecular Property Prediction
Terms related to machine learning models for predicting ADMET, solubility, toxicity, and bioactivity. Target: CTOs and heads of preclinical development.
ADMET Prediction
The computational estimation of a drug candidate's Absorption, Distribution, Metabolism, Excretion, and Toxicity properties to predict its pharmacokinetic profile.
Quantitative Structure-Activity Relationship (QSAR)
A computational modeling method that establishes a mathematical relationship between a molecule's structural features and its biological activity or chemical property.
Molecular Fingerprinting
A technique for encoding the structural features of a molecule into a fixed-length binary or integer vector for use in machine learning and similarity searching.
Extended Connectivity Fingerprints (ECFP)
A class of circular topological fingerprints, notably ECFP4, that iteratively encodes the molecular environment of each atom up to a specified diameter to capture substructural features.
Lipinski's Rule of Five
A heuristic set of four physicochemical property guidelines evaluating molecular weight, lipophilicity, and hydrogen bonding to estimate a compound's oral bioavailability.
PAINS Filter
A set of substructural alerts used to identify Pan-Assay Interference Compounds that frequently produce false positive results in high-throughput biological screening.
LogP
The logarithm of a compound's partition coefficient between octanol and water, serving as a key quantitative measure of molecular lipophilicity.
hERG Cardiotoxicity Prediction
The in silico assessment of a compound's potential to block the human Ether-Ã -go-go-Related Gene potassium channel, a critical safety endpoint linked to cardiac arrhythmia.
AMES Mutagenicity Prediction
A computational toxicology model that predicts a compound's potential to induce genetic mutations, typically using the bacterial reverse mutation assay as a benchmark endpoint.
Blood-Brain Barrier Penetration
The prediction of a molecule's ability to cross the highly selective endothelial membrane separating circulating blood from the brain's extracellular fluid.
CYP450 Inhibition
The computational prediction of a drug candidate's potential to inhibit cytochrome P450 enzymes, a major cause of adverse drug-drug interactions.
Drug-Induced Liver Injury (DILI)
The prediction of hepatotoxicity caused by a pharmaceutical agent, a complex and leading cause of drug candidate attrition and post-market withdrawal.
Multi-Task Learning in Drug Discovery
A machine learning paradigm where a single model is trained simultaneously on multiple related biological or physicochemical endpoints to improve generalization through shared representations.
Applicability Domain
The theoretical region of chemical space within which a predictive model's estimations are reliable, defined by the structural and property-based similarity to its training data.
Uncertainty Quantification
The process of assigning a confidence interval or probability distribution to a model's prediction, distinguishing between aleatoric and epistemic uncertainty in molecular property estimation.
SHAP Values
SHapley Additive exPlanations, a game-theoretic approach used to explain the output of a machine learning model by computing the marginal contribution of each molecular feature to a prediction.
Matched Molecular Pair Analysis (MMPA)
A cheminformatics method that systematically analyzes pairs of compounds differing by a single structural transformation to derive the effect of that change on a specific property.
Activity Cliff
A pair of structurally similar molecules that exhibit a drastic difference in biological potency, representing a critical challenge and opportunity in structure-activity relationship modeling.
Alchemical Free Energy Calculation
A rigorous physics-based simulation method, such as FEP+, that computationally mutates one ligand into another to predict the relative change in binding free energy.
Plasma Protein Binding
The prediction of the fraction of a drug that binds to serum proteins like albumin, directly impacting the free, pharmacologically active concentration available for distribution.
Oral Bioavailability
The fraction of an orally administered dose that reaches systemic circulation unchanged, a composite pharmacokinetic parameter influenced by solubility, permeability, and first-pass metabolism.
PAMPA
The Parallel Artificial Membrane Permeability Assay, a high-throughput in vitro method used to model passive transcellular permeability, often predicted by in silico models.
Time-Dependent Inhibition (TDI)
A form of CYP450 inhibition where the inhibitory potency increases during a pre-incubation period, often due to the formation of a more potent metabolite or a quasi-irreversible complex.
Site of Metabolism (SOM) Prediction
The computational identification of the specific atomic positions on a drug molecule where metabolic transformation by an enzyme is most likely to occur.
Molecular Transformer
A deep learning architecture that adapts the sequence-to-sequence transformer model to translate between molecular representations, such as generating a SMILES string from a molecular graph.
SELFIES
SELF-referencIng Embedded Strings, a 100% robust molecular string representation where every syntactically valid string corresponds to a valid molecular graph, designed for generative models.
ChemBERTa
A transformer-based language model pre-trained on a large corpus of SMILES strings using masked language modeling to learn generalizable molecular representations for downstream property prediction.
DeepChem
An open-source Python library providing a high-level framework for applying deep learning to drug discovery, including standardized datasets, featurizers, and model implementations.
ADMETlab
A comprehensive web-based platform that systematically evaluates a molecule's ADMET properties using a collection of high-quality, integrated predictive models.
Conformal Prediction
A model-agnostic, distribution-free framework that produces prediction sets with a rigorous, finite-sample guarantee of coverage, providing a valid measure of confidence for molecular property predictions.
Drug-Target Interaction Prediction
Terms related to binding affinity prediction, molecular docking, and polypharmacology modeling. Target: CTOs and heads of pharmacology.
Binding Affinity
The quantitative strength of the non-covalent interaction between a single biomolecule, typically a protein, and its ligand, usually expressed via thermodynamic dissociation or inhibition constants.
Molecular Docking
A computational structure-based method that predicts the preferred orientation and conformation of a ligand when bound to a target protein to form a stable complex.
Scoring Function
A mathematical function used in molecular docking to approximate the binding free energy of a protein-ligand pose, enabling the ranking of different ligands or binding modes.
QSAR (Quantitative Structure-Activity Relationship)
A ligand-based computational modeling method that establishes a mathematical relationship between the structural chemical features of a series of compounds and their biological activity.
Drug-Target Interaction (DTI)
The specific physical binding event between a drug molecule and a cellular macromolecular target, such as a protein or nucleic acid, that initiates a pharmacological effect.
Polypharmacology
The design or propensity of a single drug molecule to interact with multiple distinct molecular targets, leading to complex therapeutic or adverse biological effects.
Virtual Screening
A computational technique used in drug discovery to search large chemical libraries in silico to identify molecular structures that are most likely to bind to a specific therapeutic target.
Free Energy Perturbation (FEP)
A rigorous alchemical free energy calculation method based on statistical mechanics that computes the change in binding free energy between two related ligands through a non-physical thermodynamic path.
MM/GBSA (Molecular Mechanics/Generalized Born Surface Area)
An end-point free energy calculation method that combines molecular mechanics energy with implicit solvation models to estimate the binding free energy of a protein-ligand complex from a few simulated snapshots.
Pharmacophore Modeling
An abstraction method that identifies the essential 3D arrangement of steric and electronic features, such as hydrogen bond donors or hydrophobic centroids, necessary for a ligand to trigger a biological response.
Protein-Ligand Interaction Fingerprint
A binary or count-based vector representation encoding the specific intermolecular contacts, such as hydrogen bonds or pi-stacking, between a protein's residues and a bound ligand.
Enrichment Factor
A retrospective performance metric that quantifies how many more active compounds are identified by a virtual screening method in a top fraction of a ranked database compared to a random selection.
Decoy Generation
The process of creating a set of presumed non-binding molecules with physical properties similar to known active ligands, used as a negative control set to validate virtual screening protocols.
Conformational Sampling
The algorithmic process of generating a diverse set of low-energy 3D shapes for a flexible ligand or protein side chain to explore the potential energy landscape during docking.
Induced-Fit Docking
A docking methodology that permits conformational changes in the protein's binding pocket side chains upon ligand binding, accounting for receptor flexibility beyond a rigid-body approximation.
Covalent Docking
A specialized docking technique designed to predict the binding pose of ligands that form a permanent chemical bond with a specific nucleophilic amino acid residue on the target protein.
Proteochemometric Modeling
A machine learning approach that simultaneously uses descriptors from both the ligand chemical space and the target protein sequence space to predict bioactivity across a large interaction matrix.
Target Fishing
A computational reverse screening process that queries a single small molecule against a database of many protein structures to identify all potential macromolecular targets and off-targets.
Negative Sampling
The critical process in machine learning for DTI prediction of selecting a representative set of non-interacting drug-target pairs to train a classifier, preventing a model from becoming biased by a lack of negative data.
DeepDTA
A foundational deep learning architecture that uses two separate convolutional neural networks to learn feature representations from raw protein sequences and drug SMILES strings for predicting binding affinity.
GraphDTA
A deep learning method that represents drug molecules as 2D molecular graphs processed by a graph neural network, combined with a protein sequence CNN, to improve binding affinity prediction accuracy.
TransformerCPI
A deep learning framework that employs a transformer architecture with a gated convolutional network to capture long-range semantic interactions between protein sequences and molecular graphs for compound-protein interaction prediction.
EquiBind
An SE(3)-equivariant geometric deep learning model that performs direct, fast molecular docking by predicting the binding pocket's key points and the ligand's docked coordinates in a single forward pass without a traditional scoring function.
DiffDock
A generative diffusion model for molecular docking that frames pose prediction as a reverse diffusion process over the translational, rotational, and torsional degrees of freedom of a ligand.
Binding Pocket Detection
A computational geometry-based or deep learning method for identifying and characterizing the concave, solvent-accessible cavities on a protein surface that are capable of binding a small molecule.
Structure-Based Virtual Screening (SBVS)
A virtual screening approach that uses the experimentally determined or predicted 3D structure of a biological target to computationally dock and rank candidate ligands from a compound library.
Ligand-Based Virtual Screening (LBVS)
A virtual screening approach that uses the chemical and structural information from one or more known active ligands to search a database for other molecules with high similarity, without requiring a target structure.
ADMET Prediction
The in silico forecasting of a compound's Absorption, Distribution, Metabolism, Excretion, and Toxicity profiles, used to eliminate candidates with poor pharmacokinetic properties early in the drug discovery pipeline.
PAINS (Pan-Assay Interference Compounds)
A class of chemical substructures known to promiscuously interfere with biological assays by mechanisms like covalent modification or redox cycling, producing false positive activity readouts.
Residence Time
The reciprocal of the dissociation rate constant, representing the average duration a ligand remains bound to its target receptor, which is often a critical determinant of in vivo drug efficacy.
Molecular Dynamics Simulation
Terms related to AI-accelerated molecular dynamics, conformer generation, and conformational sampling. Target: CTOs and computational chemistry leads.
Enhanced Sampling
A class of molecular dynamics techniques that apply external biases to accelerate the exploration of a system's free energy landscape, enabling the observation of rare events within computationally feasible timescales.
Metadynamics
An enhanced sampling method that discourages revisiting previously explored states by depositing a history-dependent Gaussian bias potential along a set of predefined collective variables, effectively filling free energy minima.
Replica Exchange MD
A parallel tempering technique that runs multiple non-interacting simulations at different temperatures or Hamiltonians and periodically attempts to exchange configurations between them to overcome energy barriers.
Umbrella Sampling
A method for calculating the potential of mean force along a reaction coordinate by imposing a harmonic restraint to sample overlapping windows and subsequently unbias the distributions using the Weighted Histogram Analysis Method (WHAM).
Alchemical Free Energy
A computational technique that calculates the free energy difference between two states by simulating a non-physical pathway of intermediate states where one molecule is gradually morphed into another.
MM/PBSA
An end-point free energy calculation method that combines molecular mechanics energies with implicit solvation models to estimate the binding free energy of a ligand to a receptor from a single trajectory.
Markov State Model
A kinetic network model that discretizes a molecular system's phase space into metastable states and estimates a transition probability matrix to describe long-timescale dynamics from many short simulations.
Coarse-Grained MD
A simulation approach that reduces computational cost by grouping atoms into pseudo-particles or beads, allowing the study of larger systems and longer timescales at the expense of atomic resolution.
Martini Force Field
A widely used coarse-grained force field that maps approximately four heavy atoms to a single interaction site, parameterized to reproduce thermodynamic properties like partitioning free energies.
Particle Mesh Ewald
An efficient algorithm for calculating long-range electrostatic interactions in periodic systems by splitting the Coulombic sum into a short-range real-space term and a long-range reciprocal-space term solved via Fast Fourier Transforms.
Langevin Dynamics
A stochastic equation of motion that simulates the effect of an implicit solvent by adding friction and random noise terms to Newton's equations, acting as a thermostat and enabling Brownian motion.
Boltzmann Generator
A deep generative model that uses normalizing flows to learn a direct, invertible mapping between a simple latent distribution and the complex Boltzmann distribution of a molecular system for efficient equilibrium sampling.
Neural Network Potential
A machine-learned interatomic potential that regresses the potential energy surface from high-level quantum mechanical data, providing ab initio accuracy at a fraction of the computational cost.
Deep Potential Molecular Dynamics
A deep learning framework that constructs neural network potentials by learning the local atomic environment descriptors from first-principles data, implemented in the DeePMD-kit software.
OpenMM
An open-source, high-performance toolkit for molecular simulation featuring a custom GPU-accelerated API that allows for the rapid implementation of novel algorithms and custom force fields.
GROMACS
A highly optimized, open-source software package for molecular dynamics simulations, primarily designed for biomolecular systems, known for its extreme computational efficiency on CPUs and GPUs.
Root Mean Square Deviation
A standard metric for quantifying the structural similarity between two superimposed atomic coordinates, calculated as the square root of the average squared distance between corresponding atoms.
Time-Lagged Independent Component Analysis
A dimensionality reduction technique that identifies the slowest relaxing degrees of freedom in a molecular trajectory by maximizing the autocorrelation of the projected coordinates at a given lag time.
Collective Variable
A low-dimensional function of a system's atomic coordinates that describes the essential slow degrees of freedom governing a specific process, such as a distance, angle, or coordination number.
Gaussian Accelerated MD
An enhanced sampling method that smoothens the potential energy surface by adding a harmonic boost potential to dihedral angles and a non-harmonic boost to the total potential, accelerating transitions between states without predefined collective variables.
Conformer Generation
The computational process of generating a diverse set of low-energy three-dimensional structures for a molecule by rotating its torsional bonds to sample the potential energy surface.
ETKDG
A knowledge-based distance geometry method for generating conformers that uses experimental torsion angle preferences and small ring corrections to produce physically realistic three-dimensional molecular structures.
SHAKE Algorithm
A constraint algorithm that resets the positions of atoms involved in rigid bonds after each integration step, allowing for a larger simulation time step by freezing the fastest vibrational degrees of freedom.
Lennard-Jones Potential
A simple mathematical model describing the non-bonded van der Waals interaction between two neutral atoms, characterized by a steep repulsive term at short range and an attractive dispersion term at longer range.
Solvation Free Energy
The change in free energy associated with transferring a solute molecule from a vacuum into a solvent, representing the reversible work required to create a cavity and establish solute-solvent interactions.
Absolute Binding Free Energy
The standard free energy change when a ligand binds to a receptor from an unbound state in solution, calculated by physically separating the ligand from the binding pocket along a defined path.
Multistate Bennett Acceptance Ratio
A statistically optimal estimator for calculating free energy differences by combining data from all intermediate alchemical states simultaneously, minimizing the statistical variance of the result.
Jarzynski Equality
A non-equilibrium work theorem that relates the exponential average of the work performed during many fast, irreversible processes to the equilibrium free energy difference between two states.
GPU-Accelerated MD
The implementation of molecular dynamics algorithms on graphics processing units to leverage massive parallelization, dramatically increasing simulation throughput for non-bonded force calculations.
Ab Initio MD
A simulation method where interatomic forces are calculated on-the-fly from electronic structure theory, typically Density Functional Theory, rather than from a pre-parameterized empirical force field.
Retrosynthesis Planning
Terms related to AI-driven retrosynthetic analysis and chemical reaction yield prediction. Target: CTOs and heads of synthetic chemistry.
Retrosynthesis
The computational process of recursively deconstructing a target molecule into simpler precursor structures to identify a viable synthetic route.
Template-Based Retrosynthesis
A retrosynthetic strategy that applies a pre-defined library of reaction rules or subgraph patterns to predict disconnections in a target molecule.
Template-Free Retrosynthesis
A retrosynthetic strategy that uses sequence-based or graph-based generative models to predict precursors without relying on an explicit, pre-extracted set of reaction rules.
Semi-Template Retrosynthesis
A hybrid approach that first identifies a reaction center using a template, then completes the synthon generation using a template-free generative model.
Forward Reaction Prediction
The computational task of predicting the major product of a chemical reaction given a set of reactants and conditions.
Synthetic Accessibility Score (SAScore)
A heuristic metric quantifying the ease of synthesizing a molecule, typically calculated based on structural complexity and fragment contributions.
Molecular Transformer
A sequence-to-sequence transformer architecture that treats reaction prediction as a SMILES-to-SMILES translation task between reactants and products.
Synthon Generation
The computational step in retrosynthesis where a disconnected bond is converted into valid, synthetically equivalent molecular fragments with appropriate leaving groups.
Retrosynthetic Tree
A hierarchical data structure representing the recursive disconnection of a target molecule into precursors, where nodes are molecules and edges are reactions.
Monte Carlo Tree Search (MCTS)
A heuristic search algorithm used in retrosynthetic planning that balances exploration of new disconnections with exploitation of known high-value routes.
AND-OR Tree Search
A search strategy for retrosynthesis where an 'AND' node requires all child reactions to succeed, and an 'OR' node requires only one child pathway to succeed.
Reaction Fingerprint
A fixed-length vector representation encoding the structural transformation occurring at the reaction center, used for reaction classification and similarity searching.
Atom Mapping
The process of establishing a one-to-one correspondence between atoms in the reactants and atoms in the products of a chemical reaction.
Reaction Center Identification
The computational task of pinpointing the specific atoms and bonds that are directly involved in bond-breaking and bond-forming during a chemical reaction.
Round-Trip Accuracy
A validation metric that measures the consistency of a model by predicting the forward product from retrosynthesized reactants and checking if it matches the original target.
Reaction Class Tokenization
A method of prepending a special token representing the reaction type to the input sequence, conditioning the model to generate context-specific predictions.
Reaction Knowledge Graph
A structured graph database that encodes chemical entities as nodes and reaction relationships as edges to support reasoning over synthetic pathways.
Convergent Synthesis
A synthetic strategy where multiple fragments are synthesized independently and then coupled together at a late stage, resulting in a shorter and more efficient linear path.
Building Block Library
A curated catalog of commercially available or in-stock compounds used as terminal nodes to stop the recursive search in retrosynthetic planning.
Cost-Aware Retrosynthesis
A planning strategy that optimizes synthetic routes not just for feasibility but also for the monetary cost of starting materials and reaction steps.
USPTO Dataset
A large, publicly available chemical reaction dataset extracted from United States patents, widely used for training and benchmarking retrosynthesis models.
Pistachio Dataset
A commercial chemical reaction database derived from patent literature, curated by NextMove Software, known for its high-quality atom mapping and reaction role labeling.
Reaction Role Labeling
The task of classifying each molecule in a reaction record as a reactant, reagent, solvent, catalyst, or product.
Transition State Prediction
The computational task of predicting the 3D geometry and energy of the highest-energy structure along the reaction coordinate connecting reactants and products.
Activation Energy Prediction
The task of predicting the energy barrier that must be overcome for a chemical reaction to proceed, directly related to the kinetic feasibility of the step.
Biocatalysis Retrosynthesis
A specialized retrosynthetic planning approach that incorporates enzymatic reaction rules to design synthetic routes using enzymes instead of traditional chemical reagents.
Generative Retrosynthesis
The application of deep generative models, such as variational autoencoders or diffusion models, to directly sample novel precursor molecules for a given target.
Reinforcement Learning for Retrosynthesis
A training paradigm where a retrosynthetic agent learns a policy to select disconnections by maximizing a cumulative reward signal based on pathway quality.
Multi-Objective Optimization
A route scoring approach that simultaneously balances competing objectives like step count, yield, cost, and waste to identify Pareto-optimal synthetic pathways.
Graphormer
A transformer architecture that integrates graph structural information directly into the attention mechanism, achieving state-of-the-art results on molecular graph prediction tasks.
Protein Language Models
Terms related to transformer architectures for protein sequence representation, variant effect prediction, and enzyme function prediction. Target: CTOs and heads of protein engineering.
Protein Language Model
A transformer-based deep learning model trained on massive protein sequence databases to learn the underlying grammar, structure, and function of proteins for representation learning and generative design.
ESM-2
Meta AI's Evolutionary Scale Modeling transformer that uses masked language modeling on millions of protein sequences to generate embeddings and predict structure, function, and variant effects with state-of-the-art accuracy.
ProtBERT
A BERT-based protein language model pre-trained on UniRef100 sequences using masked language modeling to capture contextual amino acid representations for downstream prediction tasks.
ProtGPT2
An autoregressive generative protein language model based on GPT-2 that produces novel, structurally plausible protein sequences with desired functional properties.
ProGen2
A suite of large-scale autoregressive protein language models trained on over one billion sequences with conditioning tags to generate proteins across diverse families and functions.
ProteinMPNN
A message-passing neural network for inverse protein folding that predicts amino acid sequences given a target backbone structure, enabling robust de novo protein design.
Inverse Folding
The computational task of predicting an amino acid sequence that will fold into a specified three-dimensional protein backbone structure.
Zero-shot Variant Effect Prediction
The use of protein language models to score the functional impact of mutations by comparing the likelihood of the wild-type sequence against the mutated sequence without any task-specific training data.
Fitness Landscape
A conceptual mapping of all possible protein sequences to their associated biological fitness or functional activity, used to visualize evolutionary trajectories and guide engineering.
Protein Embedding
A dense, fixed-length vector representation of a protein sequence or residue learned by a language model that captures structural, functional, and evolutionary information.
Contact Prediction
The task of predicting which pairs of amino acid residues in a protein sequence are in spatial proximity within the folded three-dimensional structure.
Secondary Structure Prediction
The computational assignment of local structural motifs, such as alpha-helices and beta-sheets, to each residue in a protein sequence.
Gene Ontology Term Prediction
The automated annotation of a protein's molecular function, biological process, and cellular component using standardized Gene Ontology labels derived from sequence-based models.
Enzyme Commission Number Prediction
The computational classification of an enzyme's catalytic function by predicting its four-digit Enzyme Commission number directly from its amino acid sequence.
Subcellular Localization Prediction
The task of determining the specific compartment or organelle within a cell where a protein resides and functions, based on its sequence signals.
Thermostability Prediction
The computational estimation of a protein's ability to retain its folded structure and function at elevated temperatures, critical for industrial enzyme engineering.
Solubility Prediction
The machine learning task of forecasting a protein's propensity to remain in solution rather than aggregate, a key developability criterion for biologic drugs.
Deep Mutational Scan
A high-throughput experimental method that assays the functional effect of thousands of single amino acid substitutions across a protein, generating rich training data for variant effect predictors.
Multiple Sequence Alignment (MSA)
A computational alignment of three or more evolutionarily related protein sequences used to identify conserved regions and inform structural and functional predictions.
Position-Specific Scoring Matrix (PSSM)
A matrix representing the frequency or probability of each amino acid at every position in a multiple sequence alignment, used as an evolutionary profile for a protein family.
BLOSUM Substitution Matrix
A pre-computed matrix of log-odds scores for amino acid substitutions derived from conserved blocks of aligned protein sequences, used to score sequence similarity.
Perplexity Scoring
A metric derived from language models that quantifies how surprising or unlikely a given amino acid sequence is under the model's learned distribution, used to assess sequence quality and variant effects.
Semantic Mutagenesis
The process of navigating a protein language model's latent space to generate novel sequences with altered properties by interpolating or perturbing learned representations.
Sequence Recovery Rate
The percentage of native amino acid residues correctly predicted by an inverse folding model, serving as a standard benchmark metric for protein design accuracy.
UniRef Clusters
Clustered sets of protein sequences from UniProt that group sequences at varying identity thresholds to reduce redundancy and provide a non-redundant training set for language models.
Pfam Domain
A curated family of evolutionarily related protein regions defined by a profile hidden Markov model, representing a conserved functional or structural unit.
Byte Pair Encoding for Proteins
A subword tokenization algorithm adapted for amino acid sequences that segments proteins into frequent multi-residue tokens, enabling more efficient vocabulary representation in language models.
Masked Language Modeling for Proteins
A self-supervised pre-training objective where random amino acids in a sequence are masked and the model learns to predict them from the surrounding context, analogous to BERT training.
Autoregressive Protein Decoding
A generative modeling approach where protein sequences are produced one amino acid at a time, with each residue conditioned on all previously generated residues.
Conformational B-cell Epitope Prediction
The computational identification of discontinuous amino acid patches on a folded protein surface that are recognized by antibodies, critical for vaccine and therapeutic design.
Antibody Design and Optimization
Terms related to AI-driven antibody engineering, epitope mapping, and peptide therapeutics design. Target: CTOs and heads of biologics discovery.
Antibody Humanization
The computational process of grafting murine complementarity-determining regions (CDRs) onto human framework regions to reduce immunogenicity while maintaining antigen affinity.
Immunogenicity Prediction
The use of machine learning models to forecast the likelihood that a therapeutic antibody will provoke an unwanted immune response, primarily by identifying T-cell epitopes within the sequence.
Developability Assessment
A multi-parameter computational evaluation of an antibody candidate's biophysical properties, including solubility, stability, and aggregation propensity, to predict manufacturing and formulation risks.
Post-Translational Modification (PTM) Prediction
The in silico identification of sequence motifs susceptible to chemical or enzymatic modifications, such as deamidation or oxidation, that can compromise antibody stability and efficacy.
Antibody-Drug Conjugate (ADC) Design
The computational engineering of targeted cancer therapies where a cytotoxic payload is chemically linked to a monoclonal antibody, requiring precise prediction of conjugation sites and linker stability.
Bispecific Antibody Engineering
The design of engineered antibodies capable of simultaneously binding two distinct epitopes or antigens, often requiring computational solutions for correct heavy-chain and light-chain pairing.
Epitope Mapping
The computational identification of the specific amino acid residues on an antigen that are recognized and bound by the paratope of an antibody.
Antibody-Antigen Docking
A physics-based or deep learning simulation that predicts the three-dimensional binding pose and orientation of an antibody relative to its target antigen.
Antibody Affinity Maturation
The machine learning-guided process of iteratively introducing mutations into an antibody's CDR loops to enhance its binding strength and specificity for a target antigen.
Immune Repertoire Sequencing
The use of next-generation sequencing to profile the vast diversity of B-cell receptors in an organism, providing a data-rich source for AI-driven antibody discovery.
Single-Cell BCR Sequencing
A high-resolution technique that captures the paired heavy- and light-chain sequences from individual B cells, enabling the direct interrogation of native antibody pairing for computational analysis.
Antibody Structure Prediction
The de novo computational generation of an antibody's three-dimensional structure from its amino acid sequence, with a specific focus on accurately modeling the hypervariable CDR-H3 loop.
IgFold
A specialized deep learning model that performs rapid, template-free prediction of antibody variable domain structures directly from sequence data.
Antibody Language Model
A transformer-based neural network pre-trained on vast repositories of antibody sequences to learn the underlying grammar of immune receptors for tasks like variant effect prediction and sequence generation.
Antibody Escape Mutation Prediction
The computational forecasting of specific viral mutations that would allow a pathogen to evade neutralization by a given therapeutic antibody, crucial for assessing antiviral durability.
Antibody Pharmacokinetics (PK) Prediction
The use of machine learning to model the absorption, distribution, metabolism, and excretion profile of an antibody therapeutic, with a key focus on predicting half-life extension.
FcRn Binding Affinity Prediction
The computational estimation of an antibody's pH-dependent binding strength to the neonatal Fc receptor, the primary mechanism governing the long circulatory half-life of IgG antibodies.
Antibody-Dependent Cellular Cytotoxicity (ADCC) Prediction
The in silico modeling of an antibody's ability to engage Fc-gamma receptors on immune effector cells to trigger the killing of target cells, a critical mechanism of action for many cancer therapies.
Peptide-MHC Binding Prediction
A foundational computational immunology task that predicts the binding affinity of a peptide fragment to a major histocompatibility complex molecule, essential for T-cell epitope identification.
Generative Antibody Design
The application of generative models, such as diffusion models, to create entirely novel antibody sequences and structures with desired binding properties, moving beyond natural repertoire screening.
Antibody Co-Design
A simultaneous generative process that designs the antibody sequence and its corresponding 3D structure in a coupled fashion, ensuring the generated sequence is physically plausible and foldable.
Antibody Developability Profiling
A comprehensive computational screening cascade that aggregates predictions for aggregation propensity, thermal stability, and chemical liabilities to rank antibody candidates for manufacturing suitability.
Antibody Somatic Hypermutation Analysis
The computational tracing of point mutations accumulated in antibody variable genes during affinity maturation, used to infer lineage relationships and identify key affinity-conferring residues.
Antibody Molecular Dynamics Simulation
A physics-based computational method for simulating the atomic movements and conformational flexibility of an antibody over time, often used to assess paratope dynamics and binding interface stability.
Antibody Deep Mutational Scanning
A high-throughput experimental technique that measures the functional effect of thousands of single amino acid substitutions in an antibody, generating rich datasets for training supervised variant effect predictors.
Antibody Multi-Objective Optimization
A computational framework that simultaneously optimizes an antibody sequence for multiple, often conflicting, properties such as affinity, specificity, solubility, and immunogenicity to identify Pareto-optimal designs.
Nanobody Design
The computational engineering of single-domain antibody fragments derived from camelids, focusing on optimizing the elongated CDR3 loops for binding to cryptic or concave epitopes inaccessible to conventional antibodies.
Fc Engineering
The rational or computational modification of the antibody's constant region to modulate effector functions, such as ADCC or complement-dependent cytotoxicity, or to enhance heterodimerization for bispecific formats.
Antibody pH-Dependent Binding
The engineering of an antibody's binding affinity to be conditional on the pH environment, enabling antigen release in acidic endosomes and antibody recycling, a key strategy for extending half-life.
Antibody Logic-Gated Design
The engineering of antibodies that require the simultaneous presence of two or more tumor-associated antigens to activate, creating a Boolean AND-gate that enhances tumor specificity and reduces on-target, off-tumor toxicity.
Multi-Omics Data Integration
Terms related to machine learning for single-cell sequencing, genomic analysis, and metabolic pathway modeling. Target: CTOs and heads of bioinformatics.
Multi-Omics Integration
The computational process of combining and analyzing data from different 'omics' layers—such as genomics, transcriptomics, proteomics, and metabolomics—to create a unified, holistic view of a biological system.
Single-Cell RNA Sequencing (scRNA-seq)
A high-throughput sequencing technology that profiles the entire transcriptome at the resolution of an individual cell, enabling the discovery of novel cell types, states, and dynamic gene expression heterogeneity within complex tissues.
Batch Effect Correction
A computational technique used to remove non-biological, technical variation introduced by different experimental handling, sequencing platforms, or processing times, ensuring that true biological signals are not confounded during data integration.
Canonical Correlation Analysis (CCA)
A statistical method used in multi-omics integration to identify linear combinations of variables from two or more high-dimensional datasets that are maximally correlated, often serving as a basis for aligning different data modalities into a shared latent space.
Variational Autoencoder (VAE)
A generative deep learning architecture that learns a probabilistic, compressed latent representation of input data, widely adapted in single-cell biology for integrating multi-omics data, imputing missing modalities, and modeling complex cellular distributions.
Graph Neural Network (GNN)
A class of deep learning models designed to operate directly on graph-structured data, used in bioinformatics to model relationships between genes, proteins, or cells by passing messages between connected nodes to learn complex relational patterns.
Knowledge Graph Embedding
A technique that transforms the discrete entities and relations of a biological knowledge graph into a continuous, low-dimensional vector space, enabling the prediction of novel drug-target interactions or gene-disease associations through geometric operations.
Flux Balance Analysis (FBA)
A constraint-based mathematical modeling method used to simulate the steady-state flow of metabolites through a genome-scale metabolic network, predicting an organism's growth rate or the production rate of a specific bioproduct under defined conditions.
Spatial Transcriptomics
A collection of molecular profiling technologies that measure gene expression within intact tissue sections, preserving the spatial context of each transcript to map where specific cell types and molecular activities occur within a tissue's architecture.
Cell-Type Annotation
The computational process of assigning a biological identity label, such as 'T-cell' or 'neuron,' to individual cells or clusters in single-cell data by comparing their transcriptomic signatures to known reference profiles or marker gene sets.
Dimensionality Reduction
A mathematical technique for transforming high-dimensional omics data into a lower-dimensional space for visualization and noise reduction, with algorithms like PCA, t-SNE, and UMAP being essential for exploring cellular heterogeneity.
Gene Regulatory Network (GRN)
A directed graph representing the complex web of regulatory interactions where transcription factors control the expression of target genes, inferred computationally from multi-omics data to understand the logic of cellular differentiation and disease.
RNA Velocity
A computational method that predicts the future transcriptional state of individual cells by modeling the ratio of unspliced to spliced mRNA, thereby inferring a directional vector of cellular differentiation and developmental trajectories.
Trajectory Inference
A computational approach, also known as pseudotime analysis, that orders individual cells along a continuous developmental path based on their transcriptomic similarity, reconstructing dynamic biological processes like differentiation or disease progression.
Attention Mechanism
A core component of the Transformer architecture that allows a model to dynamically weigh the importance of different parts of the input data, enabling multi-omics models to learn context-specific interactions between genes, proteins, or regulatory elements.
Single-Cell Foundation Model
A large, pre-trained neural network, such as scGPT or Geneformer, trained on massive corpora of single-cell transcriptomic data using self-supervised learning to generate general-purpose cellular representations that can be fine-tuned for diverse downstream tasks.
Contrastive Learning
A self-supervised learning paradigm that trains a model to pull representations of similar data points together and push dissimilar ones apart in an embedding space, used in multi-omics to align cells or features across different modalities without paired labels.
Optimal Transport
A mathematical framework for finding the most efficient mapping between two probability distributions, applied in single-cell biology to align cells from different batches, time points, or omics modalities by minimizing a cost function based on their feature similarity.
Similarity Network Fusion (SNF)
An algorithm that integrates multiple omics data types by constructing a sample-similarity network for each data type and then iteratively fusing them into a single consensus network that captures shared and complementary information for patient subtyping.
Multi-Omics Biomarker Discovery
The application of machine learning and feature selection techniques to integrated genomic, proteomic, and metabolomic datasets to identify a robust panel of molecular signatures that predict a clinical outcome, such as drug response or disease prognosis.
Pathway Enrichment Analysis
A statistical method, including Gene Set Enrichment Analysis (GSEA), that determines whether a predefined set of genes or metabolites involved in a biological pathway is significantly overrepresented in a list of differentially expressed features from an experiment.
CITE-seq
Cellular Indexing of Transcriptomes and Epitopes by Sequencing, a multi-omics technology that simultaneously profiles the whole transcriptome and a panel of cell-surface proteins from the same single cell using oligonucleotide-conjugated antibodies.
Multi-Omics Data Imputation
The computational task of predicting missing values for an entire omics modality, such as proteomics, in cells where only another modality, like transcriptomics, was measured, enabling a complete multi-modal analysis from incomplete experimental designs.
Cell-Cell Communication
The computational inference of intercellular signaling networks by analyzing the expression of ligands by one cell type and their cognate receptors by another within spatial or single-cell transcriptomic data, revealing the tissue's coordination logic.
Spatial Deconvolution
A computational method that estimates the relative proportions of different cell types within each spatially barcoded spot of a spatial transcriptomics dataset by leveraging reference signatures from single-cell RNA sequencing data.
Federated Learning
A privacy-preserving machine learning paradigm where a model is trained collaboratively across multiple decentralized institutions holding sensitive multi-omics data without the raw data ever leaving its local repository, only sharing encrypted model updates.
Mendelian Randomization
A causal inference method that uses genetic variants as instrumental variables to assess the causal effect of a modifiable molecular exposure, such as a protein level, on a disease outcome, leveraging the random assortment of genes at conception.
Multi-Omics Factor Analysis (MOFA)
A statistical framework for the unsupervised integration of multiple omics data types that infers a low-dimensional set of latent factors capturing the principal sources of variation across the different data modalities from the same set of samples.
Genome-Scale Metabolic Model (GEM)
A comprehensive computational reconstruction of an organism's entire known metabolic network, encompassing all gene-protein-reaction associations, which serves as a platform for systems-level predictions of metabolic flux and cellular phenotypes.
Virtual Screening Acceleration
Terms related to AI-accelerated virtual screening, ligand-based and structure-based drug design, and chemical space exploration. Target: CTOs and heads of hit discovery.
Virtual Screening
A computational technique for rapidly evaluating large chemical libraries to identify molecules with a high probability of binding to a biological target, prioritizing compounds for experimental testing.
Ligand-Based Virtual Screening (LBVS)
A virtual screening approach that uses the known chemical and structural features of active ligands to identify novel compounds with similar properties, without requiring the 3D structure of the biological target.
Structure-Based Virtual Screening (SBVS)
A virtual screening approach that uses the three-dimensional structure of a biological target to computationally dock and score candidate ligands, predicting their binding mode and affinity.
Ultra-Large Virtual Screening
The application of virtual screening techniques to chemical libraries containing billions of compounds, often enabled by AI-accelerated docking and cloud computing to explore vast regions of chemical space.
Molecular Docking
A computational method that predicts the preferred orientation of one molecule to a second when bound to form a stable complex, used to model the interaction between a small molecule and a protein.
Scoring Function
A mathematical function used in molecular docking to estimate the binding affinity between a protein and a ligand by approximating the free energy of the complex.
Deep Docking
A deep learning-based methodology that accelerates structure-based virtual screening by training a neural network on a small subset of docking results to predict scores for the remaining library, enabling rapid triage of billion-scale databases.
Active Learning for Screening
An iterative machine learning paradigm where a model strategically selects the most informative unlabeled compounds for experimental or computational evaluation, maximizing hit discovery efficiency with minimal resources.
Quantitative Structure-Activity Relationship (QSAR)
A computational modeling method that establishes a mathematical relationship between the structural features of a set of chemicals and their biological activity to predict the activity of new compounds.
Molecular Fingerprinting
A technique for encoding the structural features of a molecule into a binary bit string or vector, enabling rapid similarity searching and machine learning applications in virtual screening.
Tanimoto Similarity
A widely used metric for comparing the similarity of two molecular fingerprints, calculated as the ratio of shared features to the total number of features, with values ranging from 0 (no similarity) to 1 (identical).
Pharmacophore Modeling
The computational construction of an abstract representation of the essential steric and electronic features required for a ligand to interact with a specific biological target and trigger a biological response.
Conformer Generation
The computational process of generating a set of low-energy, three-dimensional shapes that a flexible molecule can adopt, a critical preprocessing step for 3D similarity searching and molecular docking.
Free Energy Perturbation (FEP)
A rigorous, computationally intensive alchemical simulation method for calculating the relative binding free energy between two similar ligands, providing high-accuracy predictions for lead optimization.
Enrichment Factor (EF)
A metric that quantifies the performance of a virtual screening campaign by measuring how many more active compounds are found in a selected top fraction of a ranked database compared to a random selection.
Scaffold Hopping
The identification of novel chemotypes with a different core molecular scaffold that retain the biological activity of a known active compound, a key goal in drug discovery to circumvent patents or improve properties.
Pan-Assay Interference Compounds (PAINS)
A class of chemical compounds that frequently appear as false-positive hits in high-throughput screening due to their non-specific reactivity, aggregation, or interference with assay technology rather than genuine target binding.
ADMET Prediction
The in silico forecasting of a compound's Absorption, Distribution, Metabolism, Excretion, and Toxicity properties, used early in drug discovery to filter out molecules with poor pharmacokinetic or safety profiles.
Chemical Space Enumeration
The computational generation of a vast, explicit virtual library of all synthetically feasible molecules from a defined set of building blocks and reaction rules, such as the Enamine REAL Space.
DNA-Encoded Library (DEL) Screening
A technology that synthesizes and screens vast libraries of small molecules, each tagged with a unique DNA barcode, allowing for the simultaneous affinity-based selection of binders against a protein target.
Fragment-Based Screening
A drug discovery approach that screens libraries of very small, low molecular weight compounds to identify weakly binding chemical starting points, which are then grown or linked to create high-affinity leads.
Hit-to-Lead Optimization
The phase in early drug discovery where confirmed hit molecules are chemically modified to improve their potency, selectivity, and preliminary ADMET properties, transforming them into lead compounds.
Multi-Parameter Optimization (MPO)
A computational strategy for simultaneously balancing multiple, often conflicting, drug-like properties to identify compounds with an optimal overall profile for development.
Protein Flexibility
The dynamic nature of a protein's three-dimensional structure, which must be accounted for in advanced docking by using ensemble receptor conformations or induced-fit models to avoid missing cryptic pockets.
Covalent Docking
A specialized molecular docking technique that models the formation of a covalent bond between a reactive ligand warhead and a specific amino acid residue on the target protein, used for designing irreversible inhibitors.
Binding Kinetics
The study of the rates of association and dissociation of a drug-target complex, which can be more predictive of in vivo efficacy than equilibrium binding affinity and is increasingly modeled computationally.
AlphaFold Structures for Docking
The use of computationally predicted protein structures from models like AlphaFold2 as receptor inputs for structure-based virtual screening, enabling campaigns on targets lacking experimentally determined structures.
Chemical Library Design
The strategic selection or synthesis of a compound collection for screening, balancing diversity, drug-likeness, and novelty to maximize the probability of finding high-quality hits.
Activity Cliff
A pair of structurally similar molecules with a large difference in biological activity, representing a critical source of information for understanding structure-activity relationships and refining predictive models.
Matched Molecular Pair Analysis (MMPA)
A systematic cheminformatics approach that analyzes pairs of compounds differing by a single, well-defined structural transformation to derive rules linking specific chemical changes to changes in a molecular property.
Quantum Chemistry Machine Learning
Terms related to neural network potentials and ML-accelerated quantum mechanical calculations for molecular systems. Target: CTOs and heads of computational chemistry.
Neural Network Potential (NNP)
A machine learning model trained on quantum mechanical reference data to predict the potential energy and forces of an atomic system, bypassing the explicit solution of the Schrödinger equation for molecular dynamics.
Density Functional Tight Binding (DFTB)
A semi-empirical quantum mechanical method derived from a Taylor expansion of Density Functional Theory, parameterized using DFT calculations to provide a computationally efficient approximation for electronic structure.
Potential Energy Surface (PES)
A mathematical function describing the energy of a molecular system as a function of its atomic coordinates, the fundamental landscape upon which chemical reactions and molecular motions occur.
Force Field Parameterization
The process of determining the numerical constants in a classical molecular mechanics force field, increasingly performed by machine learning models trained on quantum mechanical data to achieve ab initio accuracy.
Equivariant Neural Network
A neural network architecture that guarantees its output transforms predictably under the symmetry operations of 3D space, such as rotation and translation, ensuring physical consistency in molecular predictions.
Ab Initio Molecular Dynamics (AIMD)
A simulation technique where interatomic forces are computed directly from electronic structure calculations at each time step, providing high accuracy but at significant computational cost compared to classical force fields.
Kohn-Sham Equations
A set of single-particle equations within Density Functional Theory that map the interacting many-body problem onto a system of non-interacting electrons, forming the practical foundation of modern quantum chemistry.
Exchange-Correlation Functional
The component of a Density Functional Theory calculation that approximates the complex quantum mechanical exchange and correlation energy of electrons, representing the primary source of approximation in DFT.
Basis Set
A set of mathematical functions used to represent the molecular orbitals of a system in quantum chemical calculations, where a larger basis set generally provides greater accuracy at increased computational cost.
Pseudopotential
An effective potential used in electronic structure calculations to replace the complex effects of core electrons, allowing the computation to focus solely on the chemically active valence electrons.
Coupled Cluster
A highly accurate post-Hartree-Fock quantum chemistry method that systematically accounts for electron correlation, often considered the 'gold standard' for generating reference data to train machine learning potentials.
Time-Dependent DFT (TD-DFT)
An extension of Density Functional Theory for calculating excited-state properties and electronic spectra of molecules by applying a time-dependent external electric field.
Enhanced Sampling
A class of molecular simulation techniques designed to overcome high energy barriers and accelerate the exploration of rare events, such as chemical reactions or protein folding, within accessible simulation times.
Free Energy Perturbation (FEP)
A rigorous statistical mechanics method for calculating the free energy difference between two states, widely used in drug discovery to predict relative binding affinities of ligands to a protein target.
Solvation Model
A computational method to approximate the effect of a solvent environment on a solute molecule, ranging from implicit continuum models to explicit atomistic representations of solvent molecules.
Particle Mesh Ewald (PME)
An efficient algorithm for calculating long-range electrostatic interactions in periodic systems, a standard component of molecular dynamics simulations to accurately model Coulombic forces.
Hamiltonian Prediction
A machine learning task where a model directly predicts the quantum mechanical Hamiltonian matrix of a system from its atomic structure, bypassing the self-consistent field cycle for electronic structure calculation.
Δ-Machine Learning
A learning strategy where a model is trained to predict the small difference between a low-level, inexpensive theory and a high-level, accurate theory, combining the speed of the former with the accuracy of the latter.
Smooth Overlap of Atomic Positions (SOAP)
A local atomic descriptor that encodes the chemical environment of an atom by constructing a smooth, continuous density of neighboring atoms, providing a rotationally and permutationally invariant representation for machine learning.
Atomic Cluster Expansion (ACE)
A systematic and complete framework for constructing permutationally and rotationally invariant atomic descriptors, forming the theoretical basis for a family of highly efficient and accurate machine learning interatomic potentials.
Self-Consistent Field (SCF)
An iterative computational procedure in quantum chemistry that solves for the electronic structure by repeatedly refining the electron density until the input and output are consistent within a defined convergence criterion.
Nudged Elastic Band (NEB)
A computational method for finding the minimum energy path and transition state between a known reactant and product state on a potential energy surface, crucial for calculating reaction rates.
Automatic Differentiation
A computational technique for accurately and efficiently calculating the derivatives of a function specified by a computer program, the core engine enabling the training of neural network potentials on forces and energies.
Force Matching
A training paradigm for machine learning potentials where the loss function directly compares the atomic forces predicted by the model to reference forces from quantum mechanical calculations.
Permutation Invariance
A design constraint for machine learning models of atomic systems ensuring that the prediction remains unchanged when the order of identical atoms in the input list is swapped, reflecting physical indistinguishability.
Active Learning Loop
An iterative training process where a machine learning potential identifies configurations where its prediction is uncertain, requests new quantum mechanical calculations for those configurations, and retrains to systematically improve its accuracy and robustness.
Uncertainty Quantification (UQ)
The process of assigning a confidence interval to a machine learning model's prediction, critical for assessing the reliability of a neural network potential and guiding active learning data acquisition.
QM/MM
A hybrid computational method that treats a small, chemically active region of a system with an accurate quantum mechanical method while modeling the larger, inert environment with a fast molecular mechanics force field.
Many-Body Expansion
A fragmentation approach that decomposes the total energy of a large molecular system into a sum of contributions from individual monomers, dimers, trimers, and so on, enabling high-accuracy calculations on large systems.
Machine Learning Force Field (MLFF)
A general term for an interatomic potential where the functional form is a machine learning model, such as a neural network or kernel method, trained on quantum mechanical data to achieve ab initio accuracy at classical force field speed.
Cryo-EM Data Processing
Terms related to deep learning for cryo-electron microscopy data analysis and protein dynamics analysis. Target: CTOs and heads of structural biology.
Single-Particle Analysis (SPA)
A cryo-EM technique where 2D projections of many identical macromolecules in random orientations are computationally combined to reconstruct a 3D density map.
Contrast Transfer Function (CTF)
A mathematical function describing how the electron microscope's objective lens aberrations modulate image contrast as a function of spatial frequency, requiring computational correction for accurate structure determination.
Particle Picking
The computational process of identifying and extracting individual macromolecular projection images from noisy cryo-EM micrographs, now often performed by deep learning models like Topaz or crYOLO.
2D Class Averaging
A computational step that aligns and averages similar particle projections to improve signal-to-noise ratio, revealing distinct orientation views and discarding junk particles before 3D reconstruction.
3D Reconstruction
The computational process of determining a macromolecule's three-dimensional Coulomb potential density map from its 2D projection images using algorithms like weighted back-projection or iterative refinement.
Gold-Standard Fourier Shell Correlation (FSC)
A resolution estimation method that splits the particle dataset into two independent half-sets for independent 3D reconstruction, comparing their Fourier shells to avoid overfitting and noise correlation.
Bayesian Polishing
A per-particle, beam-induced motion correction algorithm implemented in RELION that uses a Bayesian framework to model and reverse radiation damage and movement trajectories.
Maximum Likelihood Estimation (MLE)
A statistical method for 3D reconstruction that iteratively finds the structural model and orientation assignments that maximize the probability of observing the experimental particle images.
Expectation-Maximization (EM)
An iterative optimization algorithm used in cryo-EM refinement that alternates between computing the probability of orientation assignments (E-step) and updating the 3D density map (M-step).
Heterogeneous Refinement
A computational classification method that sorts particle images into structurally distinct 3D classes to resolve compositional or conformational heterogeneity within a single sample.
3D Variability Analysis (3DVA)
A method implemented in cryoSPARC using principal component analysis to model a continuous landscape of conformational motions from a cryo-EM particle stack.
CryoDRGN
A deep generative model using a variational autoencoder to reconstruct continuous conformational heterogeneity from cryo-EM images, learning a latent space of structural states.
DeepEMhancer
A deep learning-based post-processing tool that uses a convolutional neural network to perform simultaneous map sharpening and local amplitude scaling, improving interpretability of cryo-EM density maps.
ModelAngelo
An automated atomic model building program that uses a graph neural network to trace the protein backbone and assign amino acid sequences directly into cryo-EM density maps.
Cryo-ET
Cryo-Electron Tomography, a technique where a tilted sample is imaged to produce a 3D tomogram of pleomorphic structures like cells or organelles in their native state.
Subtomogram Averaging
A computational method analogous to single-particle analysis that aligns and averages 3D sub-volumes extracted from cryo-electron tomograms to achieve high-resolution structures in situ.
Missing Wedge Correction
Computational methods to compensate for the wedge-shaped region of missing Fourier space information inherent in tomographic tilt-series data due to limited tilt angles.
Dose Weighting
A computational compensation applied during movie frame alignment that optimally down-weights later frames to account for the cumulative loss of high-resolution information from radiation damage.
MotionCor2
A widely used GPU-accelerated software for aligning and averaging dose-fractionated cryo-EM movie frames to correct for global and local beam-induced specimen motion.
Denoising Autoencoder
A neural network architecture trained to reconstruct clean images from noisy inputs, applied in cryo-EM for micrograph denoising and tomogram restoration using techniques like Noise2Noise.
Equivariant Neural Network
A neural network architecture, such as a Tensor Field Network, that guarantees its output transforms predictably under 3D rotations and translations of the input, respecting the symmetries of physical space.
Molecular Dynamics Flexible Fitting (MDFF)
A method that uses molecular dynamics simulation to flexibly fit an atomic model into a cryo-EM density map by applying forces derived from the map's potential.
AlphaFold
A deep learning system developed by Google DeepMind that predicts a protein's 3D structure from its amino acid sequence with high accuracy, often used to generate initial models for cryo-EM map interpretation.
ProteinMPNN
An inverse folding neural network that designs amino acid sequences predicted to fold into a given protein backbone structure, used for validating and optimizing models built into cryo-EM density.
Map Sharpening
A post-processing step that applies a B-factor weighting to the Fourier amplitudes of a cryo-EM map to restore high-frequency detail attenuated by the imaging process and improve atomic feature visibility.
Local Resolution Estimation
A computational method that calculates a resolution value for each voxel in a cryo-EM map, identifying regions of structural flexibility or disorder.
Preferred Orientation
A common cryo-EM sample preparation artifact where macromolecules adopt a limited set of orientations at the air-water interface, leading to anisotropic resolution and reconstruction artifacts.
Direct Electron Detector (DED)
A digital camera technology, such as the Gatan K3 or Falcon 4, that directly detects electrons with high quantum efficiency and fast readout, enabling dose fractionation and super-resolution imaging.
Frame Alignment
The computational process of registering and averaging the multiple sub-frames of a dose-fractionated cryo-EM movie to correct for specimen drift and beam-induced motion.
Real-Space Refinement
An atomic model optimization method that directly minimizes the discrepancy between a model's calculated density and the experimental cryo-EM map in real space, often using gradient-driven or simulated annealing approaches.
Drug Repurposing Algorithms
Terms related to AI-driven drug repurposing, side effect prediction, and polypharmacology modeling. Target: CTOs and heads of translational medicine.
Drug Repurposing
The systematic identification of new therapeutic indications for existing, clinically approved drugs outside the scope of their original medical use.
Polypharmacology
The design or functional capacity of a single drug molecule to interact with multiple distinct biological targets simultaneously, producing a complex therapeutic or adverse effect profile.
Side Effect Prediction
The computational forecasting of adverse drug reactions by modeling the interaction between a drug's chemical structure and off-target biological pathways.
Drug-Target Interaction
The physical binding event between a chemical compound and a specific biomolecule, typically a protein, which modulates a biological process.
Transcriptomic Signature Matching
A computational approach that compares a disease's gene expression pattern against a reference database of drug-induced gene expression changes to identify compounds that reverse the disease state.
Connectivity Map (CMap)
A reference collection of genome-wide transcriptional expression data from cultured human cells treated with bioactive small molecules, used to discover functional connections between drugs, genes, and diseases.
Knowledge Graph Embedding
A machine learning technique that projects the entities and relations of a biomedical knowledge graph into a low-dimensional vector space to predict missing links, such as novel drug-disease associations.
Electronic Health Record Mining
The application of natural language processing and machine learning to unstructured clinical notes and structured patient data to identify real-world signals for drug repurposing opportunities.
Real-World Evidence (RWE)
Clinical evidence derived from the analysis of real-world data, such as electronic health records and insurance claims, regarding the usage and potential benefits or risks of a medical product.
Pharmacovigilance
The science and activities relating to the detection, assessment, understanding, and prevention of adverse effects or any other drug-related problem.
Target Deconvolution
The experimental or computational process of identifying the specific molecular target through which a biologically active compound exerts its phenotypic effect.
Mechanism of Action (MoA)
The specific biochemical interaction through which a drug substance produces its pharmacological effect, often involving binding to a target receptor or inhibiting an enzyme.
Drug Similarity Network
A graph-based representation where nodes represent drugs and edges connect compounds sharing chemical, biological, or side-effect profiles to infer shared therapeutic properties.
Matrix Factorization
A collaborative filtering technique that decomposes a sparse drug-disease association matrix into low-rank latent factor matrices to predict novel repurposing candidates.
Causal Inference
A statistical framework that goes beyond correlation to determine whether a specific drug exposure directly causes a particular therapeutic outcome, often using instrumental variable analysis.
Mendelian Randomization
An epidemiological method that uses genetic variants as instrumental variables to assess the causal effect of a modifiable drug target on a disease outcome, mimicking a randomized controlled trial.
Drug Combination Prediction
The computational identification of multi-drug regimens that produce a synergistic therapeutic effect greater than the sum of their individual effects.
Synergy Score
A quantitative metric, often derived from models like Bliss Independence or Loewe Additivity, that measures the degree of interaction between two drugs to determine if the combination is synergistic, additive, or antagonistic.
Zero-Shot Prediction
A machine learning paradigm where a model predicts drug-disease associations for completely unseen diseases or drugs without any specific training examples for that particular pair.
Multi-Task Learning
An inductive transfer approach that trains a single model simultaneously on multiple related prediction tasks, such as binding affinity and toxicity, to improve generalization through shared representations.
Contrastive Learning
A self-supervised representation learning method that pulls together augmented views of the same biological entity while pushing apart representations of different entities to learn robust drug or disease embeddings.
SHAP Values
A game-theoretic approach to explain the output of a drug repurposing model by computing the marginal contribution of each input feature to the final prediction.
BERT for SMILES
A transformer-based architecture adapted to learn contextualized molecular representations by pre-training on large corpora of SMILES strings using masked language modeling objectives.
Molecular Fingerprint
A binary or integer vector encoding the presence or absence of specific chemical substructures within a molecule, used as a fixed-length input for similarity searching and machine learning.
Quantitative Structure-Activity Relationship (QSAR)
A computational modeling method that establishes a mathematical relationship between the structural descriptors of a chemical compound and its biological activity.
Network Propagation
An algorithm that diffuses information, such as known drug targets, across a biological interaction network to prioritize disease-associated proteins in the local neighborhood for repurposing.
Link Prediction
A graph machine learning task that estimates the likelihood of a missing connection between two nodes in a heterogeneous biomedical network, such as a drug and a disease.
Inductive Matrix Completion
A matrix factorization variant that incorporates side information, like drug chemical structures and disease gene signatures, to predict associations for entities not present in the original training matrix.
Data Leakage
A critical validation error in drug repurposing where information from the test set, such as overlapping drug scaffolds, inadvertently influences the training process, leading to over-optimistic performance estimates.
DrugBank
A comprehensive, freely accessible online database containing detailed drug data with comprehensive drug target, mechanism, and interaction information, serving as a gold standard for computational drug repurposing.
RNA Structure Prediction
Terms related to deep learning models for predicting RNA secondary and tertiary structures. Target: CTOs and heads of RNA therapeutics.
RNA Secondary Structure Prediction
The computational task of determining the set of base pairs formed by hydrogen bonding within a single RNA strand, typically represented in dot-bracket notation.
RNA Tertiary Structure Prediction
The computational determination of the full three-dimensional atomic coordinates of an RNA molecule, defining the spatial arrangement of its helices, loops, and long-range interactions.
Pseudoknot Prediction
The specific computational challenge of identifying pseudoknots, a tertiary structural motif where bases within a loop pair with bases outside that loop, requiring algorithms beyond standard dynamic programming.
Minimum Free Energy (MFE)
The thermodynamic principle and algorithm that predicts the single most stable RNA secondary structure by minimizing the sum of empirically derived loop and stacking energy parameters.
Partition Function
A statistical mechanics calculation that sums the Boltzmann-weighted free energies of all possible RNA secondary structures to derive base pairing probabilities and ensemble properties.
Turner Energy Model
The standard nearest-neighbor empirical model that assigns thermodynamic parameters to RNA base pair stacks and loops, forming the basis for most free energy minimization and partition function calculations.
Dot-Bracket Notation
A standard string representation of RNA secondary structure where matching parentheses denote canonical base pairs and dots represent unpaired nucleotides, used as a training target for deep learning models.
Covariance Model
A probabilistic model, typically a stochastic context-free grammar, that captures both sequence conservation and correlated base-pair mutations within an RNA family to improve homology-based structure prediction.
Chemical Probing
An experimental technique, such as SHAPE or DMS, that measures nucleotide flexibility and solvent accessibility to generate reactivity profiles used as constraints for structure prediction algorithms.
SHAPE Reactivity
A specific chemical probing method that acylates the 2'-hydroxyl of flexible nucleotides, providing per-nucleotide data that correlates with local structural dynamics and is integrated as a pseudo-energy term.
RNA Language Model
A self-supervised transformer model, such as RNA-FM or RiNALMo, pre-trained on massive unannotated RNA sequence corpora to generate contextual nucleotide embeddings that encode evolutionary and structural information.
Geometric Deep Learning
A neural network paradigm, encompassing equivariant architectures like SE(3)-Transformers, designed to operate on 3D atomic coordinates while preserving physical symmetries like rotation and translation for structure prediction.
Equivariant Neural Network
A neural network architecture that guarantees its output transforms predictably under input rotations and translations, ensuring predicted 3D RNA structures are physically consistent regardless of coordinate frame.
End-to-End Learning
A deep learning strategy where a single model, such as AlphaFold 3 or RoseTTAFoldNA, directly maps raw RNA sequence to 3D atomic coordinates without relying on separate secondary structure or potential energy subroutines.
AlphaFold 3
A diffusion-based deep learning model developed by Google DeepMind that predicts the joint 3D structure of complexes including proteins, RNA, DNA, and ligands with dramatically improved accuracy over previous methods.
RoseTTAFoldNA
A three-track neural network architecture from the Baker Lab that simultaneously processes sequence, distance, and coordinate information to predict the structures of protein-nucleic acid complexes.
Diffusion Model
A generative deep learning framework that learns to reverse a noising process, starting from random atomic coordinates and iteratively denoising them into a valid 3D RNA structure, as used in AlphaFold 3 and RNA-Flow.
Predicted Local Distance Difference Test (pLDDT)
A per-residue confidence metric output by AlphaFold and related models that estimates the local accuracy of the predicted structure, serving as a critical filter for interpreting model reliability.
Root Mean Square Deviation (RMSD)
The standard metric for quantifying the global similarity between a predicted 3D RNA structure and an experimentally determined reference structure by calculating the average atomic distance after optimal superposition.
Template Modeling Score (TM-score)
A length-independent metric for assessing global structural similarity that is more sensitive to overall topology than RMSD, commonly used in RNA-Puzzles and CASP-RNA benchmarks.
Molecular Dynamics Simulation
A computational method that numerically integrates Newton's equations of motion for all atoms in an RNA system over time, used to refine predicted structures and sample conformational ensembles.
Coarse-Grained Model
A simplified physical representation of RNA, such as a one-bead or three-bead model, that groups atoms into larger interaction sites to enable faster conformational sampling during folding simulations.
Rosetta FARFAR2
A fragment assembly and Monte Carlo optimization algorithm within the Rosetta software suite specifically designed for de novo prediction of RNA tertiary structure from sequence.
Knowledge-Based Potential
A statistical energy function derived from the frequency of observed atomic interactions in known RNA structures, used to score and guide folding simulations toward native-like conformations.
RNA-Puzzles
A community-wide blind assessment experiment that evaluates the state-of-the-art in RNA tertiary structure prediction by challenging participants to predict unpublished crystallographic or cryo-EM structures.
CASP-RNA
The RNA-specific track of the Critical Assessment of Structure Prediction experiment, providing a standardized, biennial benchmark for comparing the performance of computational RNA structure prediction methods.
Cryo-EM Density Map
A 3D Coulomb potential map reconstructed from cryo-electron microscopy images, representing the electron scattering density of an RNA molecule and used as a target restraint for map-to-model fitting algorithms.
Leontis-Westhof Classification
A geometric ontology that systematically categorizes RNA base pairs by the interacting edges (Watson-Crick, Hoogsteen, Sugar) and the glycosidic bond orientation (cis or trans), enabling precise annotation of 3D motifs.
A-Minor Motif
A ubiquitous RNA tertiary interaction where an unpaired adenine inserts its minor groove edge into the minor groove of a neighboring Watson-Crick helix, often stabilizing ribosome and riboswitch structures.
G-Quadruplex
A stable, non-canonical RNA secondary structure formed by stacked guanine tetrads coordinated by a monovalent cation, representing a challenging prediction target for standard algorithms.
Partnered with leading AI, data, and software stack.
How We Work
Custom AI workflows for your Business
One-fit-all AI don't work for modern businesses. At Inferensys, we aim to understand your business & custom requirements; which we use to define most efficient agentic workflows, the data, and the tools for your business.
01
Review the use case
We understand the task, the users, and where AI can actually help.
Read more02
Pick the right approach
We define what needs search, automation, or product integration.
Read more03
Build the first useful version
We implement the part that proves the value first.
Read more04
Improve from there
We add the checks and visibility needed to keep it useful.
Read moreThe first call is a practical review of your use case and the right next step.
Talk to Us