EFMC-ISMC

The efforts of lead optimization projects are directed towards analogs that have favorable ADME profiles and are devoid of safety concerns whilst retaining target activity. In this work we present a novel computational platform to aid such projects by generating virtual analog libraries in physicochemical space compatible with the desired biological characteristics.

The main idea behind our approach is that many considered properties are governed by basic physicochemical parameters, such as ionization, lipophilicity, and molecular size. We have devised simple, yet accurate physicochemical models of intestinal absorption and passive permeation across the BBB, and general physicochemical rules that hold even for protein-ligand interactions (P-gp, hERG inhibitor specificity). Changing parameter values may have distinct, even opposing effects on different ADME properties, and the impact of a particular parameter may depend on the allowed variation ranges of others. Using the cumulative output of available predictive models enables us to account for the multitude of possible effects and identify regions in physicochemical space that are most likely occupied by analogs with the desired combination of ADME properties. Advanced techniques are also applied to improve the selection of substituents fitting these regions, including custom Hammett equations for estimating mutual effects of the core molecule and the modified substituent on the analog's pK_a.

The presented methods coupled with automatic analog generation in accordance with imposed physicochemical restrictions, make our software platform a valuable tool to guide drug discovery projects towards the most promising candidates.