Experimental and Mechanistic Validation of PARP1pred for Identifying Potent Leads.

Poly(adenosine diphosphate-ribose) polymerase 1 (PARP1) is a pivotal target for treating homologous recombination-deficient cancers through the mechanism of synthetic lethality. While machine learning has accelerated the identification of novel inhibitors, many models lack experimental validation and high-resolution mechanistic insights. In this study, we evaluated the predictive robustness of the PARP1pred model using a hierarchical pipeline. Initial bioactivity predictions for candidates in unseen chemical space were validated through biochemical and cellular sensitivity assays using a panel of isogenic TK6 cell lines. Subsequently, molecular docking, 100-ns molecular dynamics simulations, and molecular mechanics Poisson-Boltzmann surface area (MM-PBSA) energetic analysis were performed to provide a structural and thermodynamic rationale for the observed inhibitory potencies. The workflow successfully identified ZINC49069486 as a highly potent nanomolar lead that induced selective synthetic lethality in BRCA1-deficient cells. Crucially, the pipeline correctly classified ZINC67913374 as biologically inactive [median inhibitory concentration (IC50) > 100 μM], successfully identifying a false positive previously proposed in the literature as a potential inhibitor. MM-PBSA analysis revealed that the inactivity of ZINC67913374 was driven by an excessive polar solvation penalty that outweighed its favorable docking score. While ZINC98208034 and ZINC8793749 showed moderate enzymatic inhibition, they failed to induce a synthetic lethality response. These results confirm that the PARP1pred-driven hierarchical framework effectively prioritizes experimentally validated lead compound while filtering out deceptive computational hits, providing an accessible and robust strategy for PARP1 inhibitor discovery.