Molecular Docking

The engine that explores the conformational space of the ligand. It generates thousands of potential binding poses by systematically or stochastically rotating flexible bonds and translating the rigid body of the molecule within the binding site.

• Systematic Search: Incrementally rotates torsional angles, risking combinatorial explosion for highly flexible ligands.
• Stochastic Search: Uses random changes evaluated by a probability function, as seen in Genetic Algorithms (AutoDock) and Monte Carlo simulations.
• Molecular Dynamics: Simulates actual physical movement using Newtonian physics to cross energy barriers.

A mathematical approximation of the binding free energy (ΔG). It rapidly ranks the millions of poses generated by the search algorithm to distinguish correct binding modes from decoys.

• Force-Field Based: Calculates Van der Waals and electrostatic energies (e.g., DOCK).
• Empirical: Sums weighted terms like hydrogen bonds and hydrophobic contacts, calibrated against experimental affinities (e.g., ChemScore).
• Knowledge-Based: Derives statistical potentials from the frequency of atom-pair contacts in the Protein Data Bank (PDB) (e.g., PMF).

The process of generating a diverse set of low-energy 3D structures for the ligand prior to or during docking. A molecule's bioactivity is strictly dependent on its shape.

• Pre-Generation: Tools like OMEGA generate a library of ring conformations and minimize energy before docking begins.
• On-the-Fly: The search algorithm explores torsional degrees of freedom during the docking run itself, which is more thorough but slower.
• Entropy Cost: Penalizes ligands that must freeze many rotatable bonds to bind, reducing predicted affinity.

Post-processing steps to group similar binding modes and select the most likely physiological pose. Raw docking output often contains redundant solutions.

• RMSD Clustering: Groups poses by Root-Mean-Square Deviation of atomic positions (typically < 2.0 Å cutoff).
• Cluster Population: The most populated cluster often represents the entropically favored binding mode, not just the single top-scored pose.
• Visual Inspection: Critical for identifying unrealistic geometries like buried charges or unsatisfied hydrogen bond donors.

The computational process of generating a diverse ensemble of low-energy three-dimensional shapes that a flexible ligand can adopt. Effective sampling is critical because ligands rarely bind in their lowest-energy gas-phase conformation.

• Systematic search: Incrementally rotates rotatable bonds through discrete angles
• Stochastic methods: Uses random changes and Monte Carlo acceptance criteria
• Molecular Dynamics-based: Simulates physical motion to explore conformational space

Insufficient sampling leads to missed binding poses and false negatives in virtual screening.

The quantitative estimation of interaction strength between a drug candidate and its target, typically expressed as a dissociation constant (Kd) or inhibition constant (Ki). This goes beyond pose prediction to rank compound potency.

• Free Energy Perturbation (FEP): Rigorous alchemical pathway calculations with high accuracy
• MM-GBSA/MM-PBSA: End-point free energy methods balancing speed and precision
• Machine learning scoring: Deep learning models trained on structural interaction fingerprints

Accurate affinity prediction is the ultimate goal of computational hit-to-lead optimization.

A computational technique that rapidly evaluates large chemical libraries—often millions of compounds—to identify those most likely to bind a specific drug target. Docking is the core engine of structure-based virtual screening.

• Enrichment Factor quantifies how effectively active compounds are prioritized over decoys
• AUC-ROC evaluates classifier performance across all ranking thresholds
• Often combined with pharmacophore filtering to pre-screen libraries before docking

Virtual screening dramatically reduces the experimental burden of high-throughput screening.