Pocket-Conditioned Autoregressive Molecular Generative Algorithm-Driven Discovery of a Potent STAT3 Inhibitor for Colorectal Cancer Treatment

Abstract Transcription factor STAT3 is a key oncogenic driver in multiple cancer types, where its activation promotes the transcription of oncogenes and facilitates tumor progression. However, no STAT3 inhibitor has yet been approved for clinical applications. In this study, we developed a pocket-conditioned autoregressive molecular generative (PAMG) model that leveraged protein pocket information to optimize scaffold decorations. With the PAMG mode, a novel and potent STAT3 inhibitor, 11b, was discovered. In vitro, 11b selectively inhibited STAT3 phosphorylation and colorectal cancer cell proliferation at 0.5−2 μM. In an HCT-116 xenograft model, 11b (20 mg/kg, i.v.) achieved enhanced tumor growth inhibition (TGI = 81%) compared to TTI-101. Oral administration of 11b (150 mg/kg) also potently suppressed tumor growth (TGI = 86%) and synergized with the chemotherapy agent. Collectively, this study demonstrated the value of the PAMG model in drug scaffold decorations and identified 11b as a novel and potent STAT3 inhibitor.