Secondary protein structure prediction combining protein structural class, relative surface accessibility, and contact number

With huge amounts of molecular data produced from ever-increasing numbers of genomic and proteomic studies, predicting the secondary structure of proteins from amino acid sequences has become a common expectation among scientists. Several studies in the literature have demonstrated that the accuracy of such predictions can be drastically improved by incorporating additional types of protein data into the prediction process; however, no work has studied the effect of incorporating multiple types of protein data simultaneously. In this work, we report our findings from an extensive experimental study that uses neural networks designed to study the effect of using different combinations of protein data on the accuracy of predicting secondary protein structures. Overall, our experimental results indicate that accuracy improves the most when incorporating contact number, relative surface accessibility or any combination that includes at least one of the two into the prediction process.