AI-driven transformer-guided fragment-based discovery of Prostaglandin A1 as a multi-target inhibitor of PSAT1, PARP1, and PIK3CA in ovarian cancer.

Ovarian cancer remains a major clinical challenge because of pathway redundancy and therapeutic resistance, which limit the effectiveness of drugs targeting a single target. Multi-target inhibitors can simultaneously modulate key oncogenic proteins such as PSAT1, PARP1, and PIK3CA. However, identifying multi-targeted compounds remains difficult due to limitations in conventional screening approaches and chemical library diversity. In this study, we developed an integrated computational framework that combines fragment-based drug design and transformer-driven molecular screening to discover multi-target inhibitors against Ovarian Cancer. Fragment screening and optimization were performed to identify scaffold features that are compatible with all three proteins (PSAT1, PARP1, and PIK3CA). Then, in order to expand the search beyond traditional libraries, a ChemBERTa transformer model was used to perform similarity-based screening of the TCM Bank dataset, which enables prioritization of compounds aligned with fragment-derived pharmacophore characteristics. This approach identified Prostaglandin A1 as a promising multitarget candidate, demonstrating favorable predicted binding interactions with three key oncogenic proteins. Also, Prostaglandin A1 exhibited favorable WaterMap ΔG values and hydration thermodynamic profiles. The binding stability was further validated through molecular dynamics simulations followed by MM/GBSA analysis. Notably, the presence of Prostaglandin A1 in medicinal plants such as Cnidium monnieri and Allium macrostemon further supports its translational potential.