候选药物筛选中分子对接及药靶网络构建方法

Since the experimental method for drug design is time-consuming and high cost, information screening method for drug candidate molecules is an important supplement, and can explore the mechanism of action of drugs. Aiming at the receptor, this project studies the spatial recognition algorithm and energy matching measurement in molecular docking, and then obtains the appropriate conformation set targeting the protein through developing the construction method of drug targets bilayer network to the docking processing results. Firstly, considering the spatial structure characteristics of receptor and ligand, hierarchical structure encoding strategy is proposed for spatial recognition optimization algorithm. Secondly, combining with the different properties of the energy terms in the receptor-ligand matching, a reasonable multi-objective scoring function is established. Then, we will focus on the adapted number of the cluster in K-medoids algorithm that is applicable to clustering the molecular conformation set based on the spatial construction information. Finally, the drug-target double network is constructed by integrating similar networks of a variety of data sources and combing with the drug-target interactions; and then conformation set targeting the protein are given through screening and sorting the molecular conformation set with the constructed drug-target double network. This project will be significant to incorporating the new information methods such as machine learning into the molecular dynamics concepts and the biology science, which can be potentially applied to the drug discovery and design based on the structure.

药物设计的实验方法费时、成本高，候选药物分子筛选的信息方法是其重要补充，且能够探索药物作用机理。本项目针对受体靶蛋白，研究分子对接中的空间识别算法和能量匹配度量，通过发展药靶双层网络构建方法对对接结果处理，获得合适的中靶构象集。首先，充分考虑受体、配体的空间结构特点，提出分层结构编码，用于空间识别优化算法。其次，结合受体-配体匹配各能量项的不同性质，构建合理的多目标打分函数。然后，重点研究K-medoids算法中的可变簇个数问题，以适用于基于空间结构信息的分子构象集聚类。最后，融合多种数据源的相似网络，结合药物-靶标作用关系，构建药靶双层网络；并通过它筛选、排序分子构象集，获得中靶构象集。本项目对研究机器学习等信息方法与分子动力学相关概念、生物科学的结合具有重要理论意义，并在基于结构的药物发现与设计方面具有潜在的应用价值。

药物设计的实验方法费时、成本高，候选药物分子筛选的信息方法是其重要补充，且能够探索药物作用机理。本项目为求解分子对接中空间匹配这类复杂优化问题，研究合适的智能优化算法；研究了基于网络推理和基于机器学习的两类计算模型，在药靶相互作用关系预测方面的应用，研究了验证药靶互作关系的方法；提出自动确定配体分子和受体靶蛋白聚类簇数的方法，并通过优化聚类中心点更新算法，得到更适合于药物和靶标数据的聚类算法。提出了针对蛋白质的多种数据表示，研究了异质网络融合方法，以构建药靶数据的复杂特征表示；依据药物分子数据集，训练优化筛选计算模型，使之能筛选、排序候选药物数据。所设计的筛选模型还涵盖了药物分子性质预测功能。本项目对研究机器学习等信息方法与分子动力学相关概念、生物科学的结合具有重要理论意义，并在基于结构的药物发现与设计方面具有潜在的应用价值。