181 PPT7

Study Notes

Drug-Likeness and Structure-Based Virtual Screening

The "rule of five" outlines criteria for poor absorption, including molecular weight > 500, logP > 5, > 5 H-bond donors, and > 10 H-bond acceptors.
70% of "drug-like" compounds had specific ranges for H-bond donors, H-bond acceptors, rotatable bonds, and rings.
A feed-forward neural network had 92 input nodes, 5 hidden nodes, and 1 output node, correctly predicting drug-likeness in molecules.
Decision trees were used to correctly classify 91.9% of drugs but with a 34.3% false positive rate for non-drugs.
"Lead-likeness" is distinct from "drug-likeness" and involves increasing molecular complexity during lead optimization.
The "rule of three" is associated with lead-likeness and is used in fragment-based approaches to drug discovery.
The number of protein crystal structures has significantly increased, driving interest in structure-based methods for virtual screening.
Protein-ligand docking aims to predict the 3D structure of intermolecular complexes and involves exploring possible geometries and scoring poses.
The DOCK method involves constructing a negative image of the binding site using overlapping spheres and matching ligand atoms to sphere centers.
More recent algorithms for docking take ligand orientational and conformational degrees of freedom into account, using methods like Monte Carlo and genetic algorithms.
Scoring functions for protein-ligand docking aim to predict binding geometry and free energies of association, with some programs correctly predicting binding geometry in over 70% of cases.
The same function is ideally used for both docking ligands and predicting their free energies of binding.