从头预测蛋白质骨架的一种并行蚁群方法及其在CASP8/9 中的应用
摘要
Predicting the three-dimensional structures of proteins from amino acid sequences with only a few remote homologs, or <italic>de novo</italic> prediction, remains a major challenge in computational biology. The <italic>de novo</italic> modeling of the protein backbone is the prerequisite stage of a protein structure prediction process. Using a parallel ant colony optimization based on sharing one pheromone matrix, this paper proposes a parallel approach to predicte the structure of a protein backbone. The parallel approach combines various sources of energy functions and generates protein backbones with the lowest energies jointly determined by the various energy functions. All the free modeling targets in CASP8/9 are used to evaluate the performance of the method. For 13 targets in CASP8, two out of the predicted model1s selected by our approach are the best of the published CASP8 results, and seven out of the model1s are ranked in the top 10. For 29 targets in CASP9, 20 out of the best models from our predictions are ranked in the top 10, and 11 out of the model1s are ranked in the top 10. The solution described in this paper mimics the nature behavior of native protein folding by simultaneously minimizing the values of multiple energy functions. It also provides a general framework to combine different search strategies in parallel platform, which is a novel approach to solving the similar optimization problems with non-deterministic algorithms.