Abstract: It has been found that key area research during protein recognition process is typically important for proclaiming biological phenomena essence, improving drug design efficiency and decreasing cost and shortening industrial cycle of new drug design. For high complexity of protein macromolecule structure, common computer systems are not capable for continuously tracking coherence change of structure and function during protein recognition. This paper sets up a heterogeneous cluster based on GPU/CPU as a uniform biological computing environment, proposes particular data structures and algorithms to deal with large biological data, and reconfigures Kd-tree based on GPU and data access modes to improve parallel calculation performance. Thus it becomes feasible to calculate time-varying protein molecular filed fast enough to do further analysis on key area for protein recognition. The corresponding biochemistry experiment results show that the method can be used as a new effective and intuitive tool for protein interaction research.

Key words: biological computation, first principle, parallel calculation, molecular docking, protein recognition

摘要： 蛋白质识别关键区域的研究对揭示生命现象的本质规律，提高药物设计效率，降低新药物开发的成本和周期有重大的应用价值。但由于蛋白质大分子结构的高度复杂性，一般的计算机系统难以对蛋白质识别过程中结构与功能的连续性变化实现快速动态分析。设计并实现了一种基于GPU/CPU异构的集群系统，根据生物计算的特点对异构集群进行数据结构和算法设计，建立起基于GPU的Kd-tree构造和访问的高效算法，以提高系统并行计算的性能。在此基础上对蛋白质分子场的动态时变序列进行快速计算，对结果进行分类，能快速高效地确定出蛋白质的相互作用关键区域。该方法得到了相应的生化实验结果验证，为揭示蛋白质作用机制提供了一种高效直观的新方法。

关键词: 生物计算, 第一性原理, 并行计算, 分子对接, 蛋白质识别