Fig. 2From: Using protein language models for protein interaction hot spot prediction with limited dataModel training and validation using sequence, structural, and stability attributes. For each residue, we computed the aa type, conservation score, secondary structure, SASA, gas-phase energy and the respective components, as well as the polar and nonpolar solvation free energy. All 10 features were used as input for the AutoGluon training and validation procedureBack to article page