Protein Data Bank (PDB)

The archive exclusively contains experimentally determined structures, not computational predictions. The three primary methods are:

• X-ray Crystallography: The dominant method (~85% of entries), requiring protein crystallization and providing high-resolution electron density maps.
• Nuclear Magnetic Resonance (NMR) Spectroscopy: Captures proteins in solution, yielding dynamic ensembles of conformers rather than a single static model.
• Cryo-Electron Microscopy (cryo-EM): A rapidly growing method for large macromolecular complexes and membrane proteins, involving flash-freezing samples and imaging with an electron beam. Each entry is assigned a unique PDB ID, a 4-character alphanumeric code (e.g., 1BNA, 7R6G).

A PDB entry is a structured data file containing a strict hierarchy of information:

• Primary Structure: The linear amino acid or nucleotide sequence of each polymer chain.
• Secondary Structure: The assignment of local conformations, specifically alpha-helices and beta-sheets, as defined by hydrogen bonding patterns.
• Tertiary Structure: The full three-dimensional atomic coordinates (x, y, z) for all atoms in the asymmetric unit, defining the overall fold of the biological molecule.
• Quaternary Structure: The biologically active assembly, specifying how multiple individual polymer chains (subunits) non-covalently associate to form a functional complex.

Beyond atomic coordinates, each entry contains essential metadata for assessing data quality and fitness for use in downstream applications like molecular docking:

• Resolution: Measured in Ångströms (Å), this is the primary quality indicator for X-ray structures. A resolution better than 2.0 Å is considered high-quality for drug design.
• R-value and R-free: Statistical measures of how well the atomic model agrees with the experimental diffraction data. The R-free value, calculated from a test set excluded from refinement, is a cross-validation metric that guards against overfitting.
• B-factor (Temperature Factor): A quantitative measure of the atomic displacement or uncertainty for each atom, indicating flexible regions of the structure.
• Ramachandran Plot: A validation diagram showing the distribution of backbone dihedral angles (phi and psi), identifying residues with strained or disallowed geometries.

A critical distinction for computational workflows is the difference between the asymmetric unit and the biological assembly:

• Asymmetric Unit: The smallest portion of a crystal structure to which symmetry operations can be applied to generate the full crystal. It may contain only a fraction of the functional molecule.
• Biological Assembly: The functionally relevant quaternary structure, generated by applying crystallographic and non-crystallographic symmetry operations. For example, hemoglobin's biological assembly is an α₂β₂ tetramer, even if the asymmetric unit contains only a single αβ dimer. Using the incorrect assembly in structure-based drug design can lead to targeting non-physiological interfaces.

The PDB archives the structures of protein-ligand complexes, providing the ground truth data for binding affinity prediction and scoring function development. Each ligand is assigned a unique 3-letter residue name (e.g., ATP, STI for Imatinib). Key considerations for computational use include:

• Protonation States: Hydrogen atoms are typically absent in X-ray structures, requiring computational prediction of ionization states at physiological pH.
• Occupancy: A value between 0 and 1 indicating the fraction of unit cells in which a particular atom or ligand is present in a given conformation.
• Covalent Modifications: The PDB records post-translational modifications and covalently bound inhibitors, which are essential for training models that predict irreversible binding.

Computational determination of a protein's three-dimensional conformation from its amino acid sequence. Modern deep learning models like AlphaFold2 and RoseTTAFold are trained on PDB structures to predict atomic coordinates with experimental accuracy. These systems learn the complex physical and evolutionary constraints encoded in known structures, enabling structure generation for previously uncharacterized proteins in hours rather than years.

A computational method predicting the preferred orientation of a small molecule ligand when bound to a target protein to form a stable complex. Docking engines like AutoDock Vina require high-resolution PDB structures as input receptors. The process involves conformational sampling of the ligand and scoring of protein-ligand poses using physics-based or knowledge-based scoring functions derived from PDB co-crystal statistics.

The standard quantitative measure for assessing structural prediction accuracy. RMSD calculates the average distance between corresponding atoms of superimposed protein structures, typically measured in Ångströms (Å). A value below 2.0 Å is generally considered a successful prediction. This metric is the primary benchmark for evaluating docking poses against PDB co-crystal structures and for validating protein structure prediction algorithms.

The three-dimensional structural assembly formed by non-covalent binding of a small molecule within a protein's binding pocket. PDB entries containing co-crystallized ligands provide essential training data for scoring functions and binding affinity prediction models. Key features include:

• Orthosteric site: The primary active site where endogenous ligands bind
• Allosteric site: Distal pockets where binding modulates function through conformational change
• Binding pose: The specific orientation and conformation of the ligand within the pocket

Neural network architectures designed to respect the symmetries of 3D molecular structures. Equivariant neural networks guarantee that predictions remain consistent regardless of a protein's orientation in space—a critical property when processing PDB coordinate data. These models operate directly on atomic point clouds rather than 2D graph representations, preserving the rich spatial information encoded in experimentally determined structures.

A rational drug discovery paradigm that uses high-resolution PDB structures to design and optimize therapeutic candidates. The workflow includes:

• Binding site identification using geometric and energetic analysis
• Virtual screening of compound libraries against the target structure
• Free Energy Perturbation (FEP) calculations for lead optimization
• Scaffold hopping to identify novel chemotypes with improved properties This approach has yielded numerous FDA-approved drugs, including HIV protease inhibitors and kinase inhibitors for oncology.