Protein-DNA interactions play important roles in many biological processes. On the interfaces of these protein-DNA complexes, a group of hot spots are identified which contribute most to structural binding free energy. As experimental method like alanine mutation scanning to find hot spots is costly and time-consuming, many computational approaches are developed to predict hot pots and non-hot spots.
PrPDH is a web server which can efficiently predict hot spots as well as non-hot spots in protein-DNA interaction interfaces using SVM (Support Vector Machine)-based machine learning method. We totally apply 114 features with 4 types of feature groups, namely solvent accessible area, sequence, structure and network features. The 10 optimal features are selected using VSURF (Variable Selection Using Random Forests) algorithm.