A Basic Machine Learning Approach for predicting pIC50 Inhibition Values of FLT3 Tyrosine Kinase Using Quantitative Structure–Activity Relationship Modeling

Acute myeloid leukemia (AML) remains a major therapeutic hurdle, especially in patients with FLT3 tyrosine kinase mutations. The goal of this work was to create a reliable and accessible machine learning-driven quantitative structure–activity relationship (QSAR) model capable of forecasting the inhibitory activity (expressed as pIC50) of FLT3 inhibitors, overcoming the shortcomings of earlier models related to limited dataset scale, chemical diversity, and forecasting precision. A substantially expanded dataset—approximately 14-fold larger than those used in previous investigations (comprising 1350 molecules and 1269 descriptors)—was utilized to train a random forest regression model, selected for its outstanding performance and robustness against overfitting. Thorough internal assessment through leave-one-out and 10-fold cross-validation produced Q² values of 0.926 and 0.922, respectively. External testing on a separate set of 270 compounds achieved an R² of 0.941 with a standard error of 0.237.Critical molecular features governing inhibitory strength were pinpointed, enhancing understanding of the essential structural elements. Furthermore, an intuitive computational platform was built to allow quick estimation of pIC50 values and support ligand-based virtual screening, which successfully highlighted several potential FLT3 inhibitors. This study marks a notable progress in FLT3 inhibitor research by providing a dependable, practical, and streamlined method for initial drug discovery phases, with the potential to expedite the development of precision treatments for AML.