The discovery of phosphorylation-suppressed inhibitors for Signal Transducer and Activator of Transcription 3 (STAT3) presents a novel therapeutic strategy for non-small cell lung cancer (NSCLC). Despite the pivotal roles of STAT3 in cancer progression, effective inhibitors remain limited, especially for efficiently suppressing phosphorylation at Try705. This study harnesses generative deep learning to develop a model for de novo design of STAT3 inhibitors that selectively target the phosphorylated form and subsequentially induce cellular apoptosis. Initially, we constructed a generative model utilizing a generative deep learning with transfer learning and virtual screening, trained on existing STAT3 inhibitor datasets to explore the chemical space. We generated a diverse library of candidate compounds, which were subsequently screened through molecular docking and pharmacophore modeling, identifying several promising inhibitors. Compared with HG106, HG110 molecule can efficiently suppress phosphorylation of STAT3, and suppress the nucleus translocation of STAT3 in H441, which stimulated by IL6 pro-inflammatory factor. Rigorous molecular dynamics (MD) simulations were performed to evaluate the stability and interaction profiles of selected candidates within the STAT3 binding site. Among the top candidates, compounds HG106 and HG110 exhibited superior binding affinities compared to known STAT3 inhibitors. The MD simulations confirmed stable conformations and favorable interactions with key residues in the binding pocket, indicating potential for in vivo efficacy. This study demonstrates the power of generative deep learning in accelerating the identification of novel phosphorylation-suppressed STAT3 inhibitors, providing a promising direction for NSCLC therapy.
