Analysis of hot regions prediction in PPI with different amino acid mutation using machine learning algorithm

Discovering hot regions in protein–protein interaction is important for drug and protein design, while experimental identification of hot regions is a time-consuming and labor-intensive effort, meanwhile, different amino acid mutations will bring different energy changes, which lead to different features selection in predictive models; thus, the analysis of predictive models with different amino acid mutation using machine learning algorithm can be very helpful. In this paper, firstly 20 datasets are obtained according to all 20 kinds of amino acid using mutation data from the SKEMPI; then predictive models by combining feature-based classification and density-based incremental clustering were applied in datasets separately. Experiment results show that RctASA, Hydrophobicity, BpASA and BminCX are the best features to discriminate hot spots and non-hot spots at all the datasets according to different kinds of amino acid mutations, and the dataset with LEU, GLY and PRO mutations have the better prediction performance than the other amino acid mutations. These analyses help us to achieve a better insight on protein and their interactions.