MIT details CheMeleon, an open molecular model for low-data property prediction

Why it matters

CheMeleon is meant to prioritize candidates when task-specific measurements are scarce, such as ligand affinity for a newly found target. MIT says it beat classical fingerprint baselines on 58 Polaris benchmarks and saw uptake in an OpenADMET contest and at least one pharma workflow mandate. The peer-reviewed paper appeared months earlier; today’s MIT News piece is the lab’s public walkthrough of the open release. That still qualifies as a November 6 primary document for practitioners. Watch ChemProp adoption and whether pharma validates the claimed test-reduction.