ML Model OracleScreen-LILRB4 Achieves 15-Fold Enrichment for Myeloid Immune Checkpoint Binders

ML model OracleScreen-LILRB4 achieved 28.5% hit rate (15-fold enrichment) in virtual screening for LILRB4 binders, yielding 16 nanomolar compounds; lead compounds restored anti-tumor immunity in patient-derived cancer co-cultures.

• OracleScreen-LILRB4 is an ensemble ML model trained on biophysical screening data from 19,800 diverse compounds to predict continuous binding affinity (ΔFnorm) to LILRB4, achieving Spearman R=0.61, AUC=0.86.
• Prospective virtual screening of 45,760 compounds and experimental testing of 200 predictions gave a 28.5% hit rate (15-fold enrichment), with 16 nanomolar binders; lead compounds ORS-22 and ORS-14 restored T-cell function in patient-derived colorectal cancer and AML co-cultures.
• Limitations: study restricted to LILRB4 target, in vitro validation only; no in vivo or clinical data.

RadPigeon summaries are original and for information only. They are not clinical advice.