A novel method for quantitatively predicting non-covalent interactions from protein and nucleic acid sequence

Biochemical interactions between proteins and biological macromolecules are dominated by non-covalent interactions. A novel method is presented for quantitatively predicting the number of two most dominant non-covalent interactions, i.e., hydrogen bonds and van der Waals contacts, potentially forming in a hypothetical protein-nucleic acid complex from sequences using support vector machine regression models in conjunction with a hybrid feature. The hybrid feature consists of the sequence-length fraction information, conjoint triad for protein sequences and the gapped dinucleotide composition. The SVR-based models achieved excellent performance. The polarity of amino acids was also found to play a vital role in the formation of hydrogen bonds and van der Waals contacts. We have constructed a web server H-VDW (http://www.cbi.seu.edu.cn/H-VDW/H-VDW.htm) for public access to the SVR models.