Statistical Potential Based On Atom-pair Distance And-orientation And Its Application In Structure Quality Assessment

Protein structure prediction is one of the most representative and challenging topics in computational biology and bioinformatics.With the rapid development of computer technology,various structural prediction methods have emerged in the past two decades.The Critical Assessment of protein Structure Prediction(CASP)experiment,which launched in 1994 and aims to help advance the methods of identifying protein structure from sequence,has attracted scientists from different backgrounds around the world.CASP has been hailed as the Olympic contest in the field of structure prediction.According to whether the known structures are used or not,prediction methods can be divided into two categories: template-based methods and template-free(or ab initio)methods.Whatever the category,the design and development of protein energy function is the key challenge to the success of prediction methods.This thesis focuses on the potential,which is the most widely used knowledge-based energy function in the field of structural prediction.We design and construct a novel highperformance statistical potential based on the statistical information of atom-pair distance and orientation.Based on previous tests and studies,we found that atom-pair distance cutoff is the key parameter affecting potential's performance.Different distance cutoffs are adopted to optimize potential's abilities of native recognition and decoy discrimination,respectively.We also studied effects of training datasets on the performance of statistical potential by constructing different statistical potentials based on diverse non-redundant datasets filtered by different parameter sets.By applying the potential to 632 protein decoy sets and comparing it with several existing energy functions,the outstanding performance of our new statistical potential in native recognition and decoy discrimination is highlighted and discussed.