Overlap of On-demand Ultra-large Combinatorial Spaces with On-the-shelf Drug-like Libraries

Ultra-large ‘on-demand’ combinatorial libraries are revolutionizing virtual screening strategies aimed at identifying innovative hit compounds or guide fast hit to lead optimization. Due to their size (several billion molecules), these libraries are encoded as fragment spaces defined by starting building blocks and organic chemistry yielding the fully enumerated compounds, and required ad-hoc browsing similarity search algorithms. The pairwise maximum common substructure (MCS) similarity across commercial ultra-large fragment spaces has recently been addressed and shown to be surprisingly low. However, the MCS similarity to commercially available drug-like libraries (‘on-the-shelf’ physically-available compounds) remains unknown.

We therefore assembled a library of 9.3 million drug-like compounds from 25 trustable suppliers using a series of in-house druggability filters. This ‘on-the-shelf’ chemical space was next represented by a list of 2.4 million Bemis-Murcko scaffolds that were searched, using the SpaceMACS algorithm, in three on-demand fragment spaces: Enamine’s REAL (19 billion compounds), Otava’s CHEMryia (11 billion compounds), and WuXi’s GalaXi (2.1 billion compounds). Surprisingly, only the REAL fragment space significantly overlapped the commercial MCS space, suggesting that the later ‘on-demand’ fragment space is the most suitable for most hit to lead follow-up studies. The remaining two spaces (CHEMryia, GalaXi) are however interesting notably for primary hit identification among chemical spaces clearly orthogonal to that covered by commercially available screening decks. Potential biases in the current analysis will be discussed.

2022 BIOVIA Conference @AG @TL @PG