ProSPs: Protein Sites Prediction Based on Sequence Fragments

Identifying interacting sites of proteins is a relevant aspect for drug and vaccine design, and it provides clues for understanding the protein function. Although such a prediction is a problem extensively addressed in the literature, just a few approaches consider the protein sequence only. The use of the protein sequences is an important issue because the three-dimensional structure of proteins could be unknown. Moreover, such a structural determination experimentally is expensive and time-consuming, and it may contain errors due to experimentation. On the other hand, sequence based method suffers when the knowledge of sequence is incomplete.In this work, we present ProSPs, a method for predicting the protein residues considering protein sequence fragments, which are obtained using sliding windows and become the samples for an unbalance binary classification problem. We use the Random Forest classifier for data training. Each amino acid is enriched using a selected subset of physicochemical and biochemical amino acid characteristics from the AAIndex1 database. We test the framework on two classes of proteins, Antibody-Antigen and Antigen-Bound Antibody, extracted from the Protein-Protein Docking Benchmark 5.0. The obtained results evaluated in terms of the area under the ROC curve (AU-ROC) on these classes outperform the sequence-based algorithms in the literature and are comparable with the ones based on three-dimensional structure.