data2discovery

De Novo Design

· Access to >10²⁰ molecules with molecular weight <500

· Emphasis on chemically stable and reasonable molecules

· Space searchable using any computable properties

· User can influence search towards more interesting hits, away from undesirable moieties etc..

The following examples are intended to show that the representation of chemistry space produces reasonable molecules that match the desired criteria whether that be fitting in a binding site of interest or similar to known ligands. None of these molecules have been synthesized or tested.

Example 1 – A protein-protein interaction. This is an example driven by docking/scoring molecules in the binding site of the target protein. The protein structure, 2d10, is a co-crystal structure of Radixin with a helical peptide from NHERF1. NHERF1 is involved in CCR5 trafficking and as a result, could potentially be an HIV target. There are two rather deep pockets corresponding to Trp345 and Phe355 of the NHERF1 helical peptide:

The goal of this de novo design exercise was to find molecules that fit in these two pockets and explore the variety of polar interactions in and around these pockets. In all about 2 weeks of CPU time (4 processors for 3.5 days) were used to generate, dock and score 25 generations of molecules. Below are 10 example virtual hits: