Accurate Prediction of Protein Hot Spots Residues Based on Gentle AdaBoost Algorithm

Hot spots are critical for protein interactions. Since the experimental method to measure protein hot spots are very cost and time-consuming, computational method is an option to predict hot spots in protein interfaces. In this work, we use Gentle AdaBoost to identify hot spot without feature selection. For all algorithm, ASEdb and BID are used as separate training and test dataset. We extract sequential and structural information of protein, which constitutes 178 dimensional feature vectors. Comparing with other algorithms, our proposed algorithm obtained satisfactory experimental results either on training dataset or test dataset, which yeilds F1 score of 0.79 and 0.69 on training dataset and test dataset, respectively.