Tailoring Contact Based Scoring Functions for Protein Structure Prediction

Protein structure prediction (PSP) is a challenge in Bioinformatics. Given a protein’s amino acid sequence, PSP involves finding its three dimensional native structure having the minimum free energy. Unfortunately, the search space is astronomical and the energy function is not known. Many PSP search algorithms develop their own proxy energy functions known as scoring functions using predicted contacts between amino acid residue pairs where two residues are said to be in contact if their distance in the native structure is within a given threshold. Scoring functions are crucial for search guidance since they allow evaluation of the generated structures. Unfortunately, existing contact based scoring functions have not been directly compared and which one among them is the best is not known. In this paper, our goal is to evaluate a number of existing contact based scoring functions within the same PSP search framework on the same set of benchmark proteins. Moreover, we also propose a number of contact based scoring function variants. Our proposed contact based scoring functions help our search algorithm to significantly outperform existing state-of-the-art PSP search algorithm, CGLFOLD that uses contact based scoring functions. We get $$0.77\AA $$ average RMSD and 0.01 average GDT values improvement than CGLFOLD.