PPB4 predicts inactive as well as active targets for small molecules
- PPB4 is a deep neural network trained on ChEMBL35 to forecast which targets a query molecule will hit.
- Unlike its predecessor PPB3 and similar tools, it also flags likely inactive targets, and targets where the molecule resembles both active and inactive compounds.
- Each prediction comes with the nearest training-set neighbors annotated as active; two prospective case studies illustrate use, and the tool is freely available online.
💡 Points Based on the abstract
An "inactive" call is only as good as the negatives behind it, and ChEMBL negatives are sparse, assay-dependent and often just a single-concentration miss. I'd want to know how the inactive class is defined and whether it was scored against held-out confirmed inactives, not only actives. On my own project I'd first test it on known off-targets of my chemotype before trusting it to trim a counter-screen panel.