Deep Origin reports a 30.6% wet-lab hit rate

Deep Origin screened 80 billion virtual compounds and reports that 56 of 183 tested molecules inhibited CD73, a 30.6% wet-lab hit rate versus roughly 0.3% in a prior machine-learning screen. The company’s preprint announcement says DODock and DOScore combined machine learning with physical refinement, while noting the result is company-authored and not yet peer reviewed. Biotech teams should demand leakage-resistant splits and prospective assay results before buying screening software—the validation standard that matters for AI systems designed to shorten drug discovery.