An integrative machine learning, explainable AI, molecular simulation, and cytotoxicity validation framework for the discovery of selective SIRT1 inhibitors against triple negative breast cancer

Background Triple negative breast cancer (TNBC) is an aggressive disease characterized by a poor prognosis, high decline rates, lack of hormonal receptors for targeted therapy, limited effectiveness of existing treatments, and the emergence of chemoresistance. Sirtuin1 (SIRT1) is an epigenetic modifier and nicotinamide adenine dinucleotide (NAD+) dependent class III histone deacetylase (HDACs) protein. It promotes the modulation of several tumour suppressors and oncogenes. The evidence also suggests that inhibiting SIRT1 activity using selective SIRT1 inhibitors could restore E-cadherin expression and suppress EMT-mediated metastasis in TNBC cells. Though we have many SIRT1 protein inhibitors, they exhibited off-target effects, low isoform selectivity, and inefficiency at the clinical trial stage. Methods In this study, we aim to identify SIRT1 isoform inhibitors by utilizing an integrated computational approach. Three stages of ML modelling were performed to find the best model from the SIRT1-based dataset. The QDA + ROS and XGBClassifier + ROS models were identified as the most robust, and they were subjected to the SHAP framework (XAI approach) to address the “black box” nature of the developed ML models. The NPASS natural compound dataset was first screened with the applicability domain of the developed models, followed by a two-step virtual screening with UniDock and AutoDock GPU in Scientiflow. Finally, the selected compounds were taken for molecular dynamics simulation, with rigorous trajectory analysis, and preliminary experimental validation was done. Results The compounds with NPASS IDs: NPC216682, NPC480509, NPC210910, and NPC247082 were identified as the most promising hits. Among these hits, Praziquantel (NPC480509), the only available test compound with reported anticancer properties, revealed the cytotoxic nature on MDA-MD-231 and MCF7 breast cancer cell lines, whereas non-cytotoxic on normal breast cell line (MCF10A). Conclusion This study is exploratory. Exact SIRT1 selective inhibition by Praziquantel, along with other hits, may be further studies thorugh in-vitro and in vivo evaluation to understand the exact mechanism of action of these hits. The integrated in silico, and preliminary in-vitro approach proposed in this study could lead to innovative outcomes when applied in a pharmaceutical framework distinct from traditional methods.