Applying Feature-Based Resampling to Protein Structure Prediction

Abstract

Protein structure prediction (PSP) is an important task as the three-dimensional structure of a protein dictates what function it performs. PSP can be modelled on computers by searching for the global free energy minimum based on Afinsen's 'Thermodynamic Hypothesis'. To explore this free energy landscape Monte Carlo (MC) based search algorithms have been heavily utilised in the literature. However, evolutionary search approaches, like Genetic Algorithms (GA), have shown a lot of potential in low-resolution models to produce more accurate predictions. In this paper we have evaluated a GA feature-based resampling approach, which uses a heavy-atom based model, by selecting 17 random CASP 8 sequences and evaluating it against two different MC approaches. Our results indicate that our GA improves both its root mean square deviation (RMSD) and template modelling score (TM-Score). From our analysis we can conclude that by combining feature-based resampling with Genetic Algorithms we can create structures with more native-like features due to the use of crossover and mutation operators, which is supported by the low RMSD values we obtained.