Absolute Quality Assessment of Protein Models

Abstract

Modeling methods increasingly attempt to close the gap between the number of known protein-coding sequences to that of structurally resolved proteins, but the results of these methods have been of mixed success. Adequate models can be built for proteins with high sequence similarity to a structurally resolved protein and occasionally modeling even succeeds in the absence of a good template, but currently no method exists to reliably rate the quality of the models.Many protein structure prediction methods rate protein models using an established scoring function by comparing the energies of an ensemble of structures and choosing the lowest energy members of said ensemble as the prediction. The acceptance of theoretical protein models is limited in the life-sciences, as currently no method exists to rate the quality of a protein model a-priori, i.e. from the model alone.Here we investigate an approach to provide an a-priori estimator of the quality of a protein model using a free-energy scoring function[1], without comparing it to a competing ensemble. We devised a N-dimensional statistical test based on the per-residue energies of amino acids in a set of high-resolution experimental structures. The quality of the protein structures can be assessed by comparison against these statistics. We were able to discriminate the low quality models for 93% of the 160 proteins tested, which is increased further to 99%, when excluding proteins, which bind cofactors or DNA; interactions not considered by the energy model or the training set. In combination with bioinformatics based methods that exclude proteins that are not covered by the scoring function, this measure for quality assessment of protein models may help increase the acceptance of qualified theoretical protein models in the life-sciences.[1] Verma, A.,Wenzel, W. (2009). A Free-Energy Approach for All-Atom Protein Simulation. Biophys.J 96,3483-3494.