What happened
Researchers combined multiple machine learning models trained on different molecular representations to better predict which compounds will work as drugs, using a metric (early hit enrichment) that matters for actual lab testing rather than abstract performance scores. This makes it faster and cheaper to filter through millions of candidate molecules and find the promising ones worth testing in the lab.
Why it matters
Drug discovery screening libraries contain millions of molecules; testing even a tiny fraction experimentally is expensive. If a machine learning model can reliably rank which compounds are most likely to work before any wet lab testing, it compresses the funnel earlier and cuts screening costs — but only if the model's confidence in its top predictions (not overall accuracy) is actually reliable.