Discriminate protein decoys from native by using a scoring function based on ubiquitous Phi and Psi angles computed for all atom Avdesh Mishra, Sumaiya Iqbal, Md Tamjidul Hoque n Computer Science, University of New Orleans New Orleans, LA 70148, USA HIGHLIGHTS  Ubiquitous dihedral angles (uD) are mined to generate energy compo- nents.  Regularized exact regression based predicted ASA is modeled into energy score.  Weight-optimized linear sum of core, ASA and uD energies formed 3DIGARS3.0.  The new Energy function 3DIGARS3.0, outperforms state-of-the-art methods. GRAPHICAL ABSTRACT article info Article history: Received 26 September 2015 Received in revised form 26 February 2016 Accepted 17 March 2016 Available online 28 March 2016 Keywords: Protein structure prediction Decoy structure Native structure Genetic Algorithm Optimization abstract The success of solving the protein folding and structure prediction problems in molecular and structural biology relies on an accurate energy function. With the rapid advancement in the computational biology and bioinformatics fields, there is a growing need of solving unknown fold and structure faster and thus an accurate energy function is indispensable. To address this need, we develop a new potential function, namely 3DIGARS3.0, which is a linearly weighted combination of 3DIGARS, mined accessible surface area (ASA) and ubiquitously computed Phi (uPhi) and Psi (uPsi) energies – optimized by a Genetic Algorithm (GA). We use a dataset of 4332 protein-structures to generate uPhi and uPsi based score libraries to be used within the core 3DIGARS method. The optimized weight of each component is obtained by applying Genetic Algorithm based optimization on three challenging decoy sets. The improved 3DIGARS3.0 out- performed state-of-the-art methods significantly based on a set of independent test datasets. Published by Elsevier Ltd. 1. Introduction Energy function is one of the most important component of protein folding and structure prediction problem. We need an accurate energy function that can assign the lowest global or, local energy to the native protein. Although, there exist fields of quan- tum mechanics (Cornell et al., 1995; Brooks et al., 1983), that can solve the existing problem, the equations are tedious to work with as the scope of the space or domain is fairly complex. Thus, statistical-based, empirical or knowledge-based energy functions have been developed (Samudrala and Moult, 1997; Zhou and Zhou, 2002; Tanaka and Scheraga, 1976; Jernigan and Bahar, 1996; Kor- etke et al., 1996; Tobi and Elber, 2000) and found to be more successful than potentials based on quantum mechanics. One of the major reason for the success of knowledge-based potential is the growing number of experimental protein structures. The knowledge of 3D (three-dimensional) structures of the target proteins and their binding sites with ligands is important for rational drug design. Although, X-ray crystallography is a powerful tool in this regard, it is time-consuming and expensive, and not all proteins can be successfully crystallized. Membrane proteins are difficult to crystallize and most of them will not Contents lists available at ScienceDirect journal homepage: www.elsevier.com/locate/yjtbi Journal of Theoretical Biology http://dx.doi.org/10.1016/j.jtbi.2016.03.029 0022-5193/Published by Elsevier Ltd. n Corresponding author. E-mail addresses: amishra2@uno.edu (A. Mishra), siqbal1@uno.edu (S. Iqbal), thoque@uno.edu (M.T. Hoque). Journal of Theoretical Biology 398 (2016) 112–121