High-throughput methods for computer-aided drug design pertaining to flexibility, selectivity and lipophilicity.

摘要

This dissertation describes advancements and applications of high-throughput computational techniques for drug discovery and development. Particularly, emphasis has been placed on issues pertaining to the discovery of compounds with broad and narrow specificity, using multiple protein structure and binding sites for docking, as well as prediction of membrane solubility of possible drug candidates.An allosteric binding site of HIV-1 reverse transcriptase shows significant important flexibility through rearrangement of side chains in the site upon binding a ligand. Virtual screening studies were carried out on multiple X-ray structures, and results were used in consensus to discover new inhibitors with low micromolar inhibition. Studies reported here demonstrate the viable use of multiple structures in docking, which can be used to address broad specificity in the case of resistance conferring mutations.Furthermore, it may be the case that narrow specificity is desired when comparing different binding sites. Atomic level resolution was required in the case of targeting Plasmodium falciparum TS-DHFR after experimental detail was inconclusive about where an active compound was binding. Approximate relative free energies of binding were calculated using molecular mechanics and a generalized Born implicit solvent model (MM-GB/SA), to show that inhibitory compounds were binding to the active site. X-ray crystal structures verified these computational findings.Beyond enzyme inhibition, properties pertaining to drug delivery and toxicity must be considered early in preclinical development. Ligand lipophilicity, estimated as the chloroform/water partition coefficient (log Pc/w) and associated with membrane solubility, was predicted using a generalized Born implicit solvent model of chloroform. The model was trained on a set of 107 small organic molecules to reproduce free energies of solvation. The calculated data was successfully correlated to experimental data with an R 2 of 0.72, and a mean unsigned error of 0.78 kcal/mol for free energy of solvation. Additionally, prediction of log Pc/w for 30 molecules resulted in a correlation with an R2 of 0.92 and a mean unsigned error of 0.44 log units for experimental and calculated data.