A Simple Machine Learning-Based Quantitative Structure-Activity Relationship Model for Predicting pIC Inhibition Values of FLT3 Tyrosine Kinase.

Acute myeloid leukemia (AML) presents significant therapeutic challenges, particularly in cases driven by mutations in the FLT3 tyrosine kinase. This study aimed to develop a robust and user-friendly machine learning-based quantitative structure-activity relationship (QSAR) model to predict the inhibitory potency (pIC values) of FLT3 inhibitors, addressing the limitations of previous models in dataset size, diversity, and predictive accuracy. Using a dataset which was 14 times larger than those employed in prior studies (1350 compounds with 1269 molecular descriptors), we trained a random forest regressor, chosen due to its superior predictive performance and resistance to overfitting. Rigorous internal validation via leave-one-out and 10-fold cross-validation yielded Q2 values of 0.926 and 0.922, respectively, while external validation on 270 independent compounds resulted in an R value of 0.941 with a standard deviation of 0.237. Key molecular descriptors influencing the inhibitor potency were identified, thereby improving the interpretability of structural requirements. Additionally, a user-friendly computational tool was developed to enable rapid prediction of pIC values and facilitate ligand-based virtual screening, leading to the identification of promising FLT3 inhibitors. These results represent a significant advancement in the field of FLT3 inhibitor discovery, offering a reliable, practical, and efficient approach for early-stage drug development, potentially accelerating the creation of targeted therapies for AML.