A Proteochemometric Model for Ligands of the SLC5 Transporter Family.

The SLC5 family of solute carriers is of significant interest for drug development due to its role in many disease processes. Building on the recent elucidation of SGLT2's structure, we developed a proteochemometric model for SLC5 inhibitors in order to gain information on selectivity-driving amino acids in the binding site. Ensemble-based algorithms, namely random forest (RF) and gradient-boosted trees, proved the best suited for the task reaching high accuracy in both activity and selectivity predictions with Morgan circular fingerprints and Z-scales for ligand and protein features, respectively. Inclusion of protein sequence as input parameters for the PCM modeling allowed identification of Leu286 in hSLGT2 as a new potential key binding site residue crucial for selectivity. Furthermore, the PCM model also performed well in predicting the effect of single-point mutations at hSGLT2 on the binding affinity of empagliflozin. The obtained models are available in the form of a Jupyter notebook.