Predicting protein sidechain conformation with statistical potential

【作者】 戚逸飞； 陈浩； 来鲁华；

【Author】 Qi Yifei~2 Chen Hao~(1,2) Lai Luhua 1 Beijing National Laboratory for Molecular Sciences,State Key Laboratory for Structural Chemistry of Unstable and Stable Species,College of Chemistry and Molecular Engineering and 2 Center for Theoretical Biology,Peking University,Beijing 100871,China

【机构】 Center for Theoretical Biology,Peking University； Beijing National Laboratory for Molecular Sciences,State Key Laboratory for Structural Chemistry of Unstable and Stable Species,Co liege of Chemistry and Molecular Engineering；

【摘要】 <正>Prediction of protein side chain conformation is a key step in protein design,as well as protein structure prediction and homology modeling.We have developed a method for protein side chain conformation prediction using a combination of statistical potential and force field.Five different statistical potentials,namely DFIRE,RAPDF,KBP,FRCBP and SACBP,in combination with two different van der Walls energies and a rotamer probability term were tested in the calculations. When using softened OPSL-AA van de Walls energy,RAPDF gave the best result,while SACBP and linear VDW energy achieved the highest accuracy over all.The program was run over a dataset of 180 proteins with 34342 side chains and compared with SCWRL3,a popular sidechain prediction program.The total x₁ and x₁₊₂ dihedral angle accuracies are 83. 15% and 74.06% using an iterative optimization method and 83.75% and 74.76% using Monte Carlo simulated annealing optimization,a slight improvement over the SCWRL3 result:82.50% and 73.00%,respectively.In addition,our program also runs 13.6% faster than SCWRL3,which has some advantages in applications to protein design.Modification of this program to enable full sequence design is underway.更多还原

【Abstract】 Prediction of protein side chain conformation is a key step in protein design,as well as protein structure prediction and homology modeling.We have developed a method for protein side chain conformation prediction using a combination of statistical potential and force field.Five different statistical potentials,namely DFIRE,RAPDF,KBP,FRCBP and SACBP,in combination with two different van der Walls energies and a rotamer probability term were tested in the calculations. When using softened OPSL-AA van de Walls energy,RAPDF gave the best result,while SACBP and linear VDW energy achieved the highest accuracy over all.The program was run over a dataset of 180 proteins with 34342 side chains and compared with SCWRL3,a popular sidechain prediction program.The total x₁ and x₁₊₂ dihedral angle accuracies are 83. 15% and 74.06% using an iterative optimization method and 83.75% and 74.76% using Monte Carlo simulated annealing optimization,a slight improvement over the SCWRL3 result:82.50% and 73.00%,respectively.In addition,our program also runs 13.6% faster than SCWRL3,which has some advantages in applications to protein design.Modification of this program to enable full sequence design is underway.更多还原