in this work, an original methodology to eliminate as many molecules
as possible from proprietary in house database after a Virtual Library
Screening (VLS) using TRIPOS FlexX
for docking and the TRIPOS
Cscore module for scoring.
a ranked-based consensus scoring function based
on the five scoring functions : FlexX Score, G_Score,
D_Score, ChemScore, and PMF
Score available in TRIPOS Cscore module.
We have selected
a “diverse” database (1420 compounds of the NCI Diversity
database) and a set of “diverse” protein-ligand complexes
to train a neural network. Among the FlexX 200 protein-ligand dataset
complexes, FlexX was able to generate 78 complexes with an accurate
RMSD (<2 Å). We have then hidden each of these 78 ligands
into the Diversity database and High Throughput docked each of them
for each of the 78 targets.
We trained a neural network on the scoring results
to combine the 5 scoring functions and generate our Generalist Function