Skip to content
DIMECHAIN

Drug discovery

Accelerated Drug ScreeningMaturity: concept

Virtual screening has to be cheap enough to run across a whole library and accurate enough not to throw away good molecules. Docking scores track binding affinity loosely, so shortlists carry false positives you can measure and false negatives nobody ever finds out about. We are working on scoring for the middle of the funnel, tested by whether it ranks known binders highly without being told.

concept An idea we find credible. Nothing has been built or measured.

What this is

The problem

Virtual screening filters enormous compound libraries down to a shortlist worth synthesising. The filter's job is to be cheap and to avoid discarding good candidates.

Where the current approach strains

Docking scores correlate with binding affinity loosely enough that shortlists carry many false positives, and — more expensively — unknown false negatives. Nobody finds out about a good molecule that was screened out.

What we are exploring

Better scoring for the middle of the funnel, where cheap methods lose reliability and accurate methods are unaffordable at scale.

What would have to be true

Retrospective screening against libraries with known actives: does the method rank the known binders highly without having been told? Enrichment on held-out actives is the standard test and the right one.

Where it applies

Related

Want to co-develop this?