Fragment-Based Generation

The foundation of fragment-based generation relies on curated libraries of low-molecular-weight compounds (typically < 300 Da) with high aqueous solubility. These fragments exhibit high ligand efficiency—strong binding affinity per heavy atom—making them ideal starting points. Libraries are often filtered using the Rule of Three: molecular weight < 300, hydrogen bond donors ≤ 3, hydrogen bond acceptors ≤ 3, and cLogP ≤ 3. Computational enumeration of fragment libraries enables systematic exploration of chemical space while maintaining drug-like properties.

Fragment linking computationally connects two or more fragments that bind to adjacent subsites within a protein pocket. The algorithm must solve a constrained molecular generation problem: identifying linker moieties that maintain optimal geometry, preserve binding poses, and satisfy synthetic accessibility. Key approaches include:

• Recap-based fragmentation for retrosynthetic linking rules
• Shape-based scoring to maintain 3D complementarity
• Bond vector alignment to ensure proper fragment orientation
• Linker enumeration from databases of synthetically accessible bridges

Fragment growing iteratively extends a single bound fragment by adding functional groups or rings to explore nearby binding pockets. This structure-based design approach uses:

• Grid-based scoring functions to evaluate atom placements
• Pharmacophoric constraints to guide growth direction
• Free energy perturbation estimates for ranking modifications
• Synthetic tractability filters to ensure viable chemistry The process balances exploration of chemical diversity with maintaining the original fragment's binding mode and efficiency.

Computational methods identify energetically favorable binding regions on the protein surface where fragments are most likely to bind. Techniques include:

• FTMap: Solvent mapping using organic probe molecules to locate consensus clusters
• GRID: Calculating interaction energies between chemical probes and the protein surface
• Mixed-solvent molecular dynamics: Simulating fragment-sized probes in explicit solvent to reveal cryptic pockets
• Machine learning hotspot predictors trained on fragment screening data These hotspots guide fragment placement and prioritization for linking or growing operations.

Fragment-based generation explicitly optimizes ligand efficiency (LE)—binding affinity normalized by molecular size—to avoid the molecular obesity common in HTS-derived leads. Key metrics tracked during generation:

• Lipophilic ligand efficiency (LLE): pIC50 - logP, penalizing excessive hydrophobicity
• Group efficiency (GE): ΔΔG per heavy atom added during growing
• Fit quality (FQ): Ratio of observed to predicted affinity based on polar atom count
• Size-independent metrics to maintain fragment-like properties in final compounds Multi-objective optimization ensures potency gains without sacrificing drug-like physicochemical profiles.

The computational pipeline for evolving fragments into lead compounds involves iterative structure-based design cycles:

1. Fragment docking to generate initial binding poses
2. Pharmacophore mapping to identify critical interaction features
3. Fragment merging when overlapping fragments suggest combined scaffolds
4. Scaffold hopping to explore alternative cores while preserving key interactions
5. Multi-parameter optimization balancing potency, selectivity, ADMET, and IP novelty Each iteration refines the molecule using experimental feedback from biophysical assays like SPR or X-ray crystallography.

The curation of a small but diverse collection of low-molecular-weight compounds (typically < 300 Da) with high aqueous solubility. A well-designed library follows the Rule of Three (MW < 300, logP ≤ 3, H-bond donors ≤ 3) and maximizes chemical diversity while ensuring synthetic tractability for downstream elaboration.

Two primary strategies for fragment elaboration:

• Fragment Linking: Connecting two non-overlapping fragments that bind to adjacent sub-pockets, often using a rigid linker to preserve binding geometry.
• Fragment Growing: Iteratively extending a single fragment into nearby pockets by adding functional groups that form new interactions. Both require precise structural data from X-ray crystallography or NMR.

Fragment hits exhibit weak binding (mM to µM range), requiring sensitive detection methods:

• Surface Plasmon Resonance (SPR): Measures real-time binding kinetics.
• NMR Spectroscopy: Detects ligand binding via chemical shift perturbations.
• Thermal Shift Assay (TSA): Monitors protein thermal stability changes upon binding.
• X-ray Crystallography: Provides atomic-resolution binding mode data for structure-guided elaboration.

Quantitative measures that normalize binding affinity by molecular size to prioritize fragments with optimal binding efficiency:

• Ligand Efficiency (LE): ΔG / heavy atom count.
• Lipophilic Ligand Efficiency (LLE): pIC50 - logP.
• Group Efficiency (GE): Measures the binding contribution of added functional groups during growing. High LE fragments maximize binding per atom, leaving room for ADMET optimization.

In silico placement of fragment libraries into a target protein's binding site using docking algorithms. Challenges include:

• Scoring function inaccuracy: Weak fragment binding energies are difficult to rank.
• Solvation effects: Displacement of water molecules can dominate binding thermodynamics.
• Ensemble docking: Using multiple protein conformations to account for receptor flexibility. Often combined with molecular dynamics to refine poses.

The systematic process of transforming a validated fragment hit into a lead compound with nanomolar potency:

1. Structure-guided growing: Adding atoms to fill adjacent pockets.
2. Property monitoring: Tracking MW, logP, and solubility to maintain drug-likeness.
3. Selectivity profiling: Screening against related targets to avoid off-target effects.
4. Synthetic tractability: Ensuring the elaborated molecule remains synthesizable. The goal is to preserve high ligand efficiency throughout optimization.