Despite long-standing interest in heat shock protein 90 (HSP90) inhibitors for oncology indications, none have yet cleared the hurdles of clinical trials to become an approved drug. Many of the existing inhibitors rely on common cores derived from resorcinol, purines, or natural products, further limiting novelty and the chances of clinical success.
To discover novel HSP90 inhibitor chemotypes, the interaction profiles of over 4 million commercially available compounds with ~8,500 proteins were rapidly predicted using Cyclica’s MatchMaker deep learning technology. By generating these ~34 trillion predictions, a rank-ordered list of compound-protein interactions was created. This allows for the selection of compounds based not only on predicted binding with the desired target HSP90, but also on the lack of interaction with targets known to be liabilities to a potential drug candidate.
Screened compounds having HSP90α as one of the top 10 most likely interactions were selected for additional analysis. After filtering for druglike properties and low similarity to existing HSP90 inhibitors (Morgan 3 similarity ≤ 0.3), the compounds were clustered by structural similarity and a diverse set of 274 compounds was selected for biochemical evaluation at a single concentration. While testing of the full compound set is still ongoing, 20 compounds have met the criteria for progression to concentration-response testing and selectivity evaluation.

Machine learning enabled discovery of novel HSP90 inhibitor chemotypes

ACS Spring Meeting 2021

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.