| Resumen |
This work is focused on implementing and evaluating the Random Forest Classifier (RFC), among other classical machine learning models, on predicting the residues at the interface of protein-protein interactions (PPI) that contribute most of the binding free energy (called hot spots and hot regions). The dataset comprises twenty-nine bone morphogenetic proteins (BMPs) complexes from the Protein Data Bank (PDB). We used just six features such as B-factor, hydrophobicity index, prevalence score, accessible surface area (ASA), conservation score, and the ground-state energy of the amino acids, which were calculated using the Density Functional Theory (DFT). Proving and testing several machine learning methods, we selected the RCF because of its better performance using classical classification metrics and tests. An optimal parameter selection of the RFC reached a better performance using this dataset with around 90 % with the correct class assigned (hot spot & hot region / non-hot spot hot region) residues. © 2023 Instituto Politecnico Nacional. All rights reserved. |
