药物潜在靶标的识别对于创新药物的研发、药物副作用评价及旧药新用等领域都有着非常重要的意义。然而受制于通量、精度和成本的限制，生物实验方法通常难以广泛开展。作为一种快速而低成本的手段，计算机辅助的药物-靶标识别方法正受到越来越多的重视。如何发展快速、精确的药物-靶标相互作用预测方法，成为信息学家和药物学家面临的新问题。本项目研究基于氨基酸序列信息及药物化合物分子指纹信息，并结合机器学习方法和模式识别理论，对药物与靶标相互作用进行预测研究。首先，研究氨基酸序列的矩阵表示及基于低秩逼近的特征编码新方法；研究基于分子指纹的药物化合物结构数值序列描述新方法，从而抽取出它们所蕴含的能定量刻画其内在本质属性的特征信息。然后，基于非均衡数据集的多分类器集成方法和基于Hadoop的分布式机器学习框架被用来快速、精确的预测药物-靶标之间的相互作用关系。本项目的研究成果将为靶向药物的设计提供理论指导。

Identifying potential drug targets is very important for modern drug discovery, drug side effects evaluation and new uses for old drugs. However, the biological experiments are usually low-throughput, inaccurate and expensive. Therefore, there is an urgent need to develop novel in silicon prediction approaches capable of identifying potential drug–target interactions in a timely manner.In this project, we aim at predicting the drug-target interactions by employing the amino acids sequence information and the fingerprints for drugs. Firstly, mapping each protein sequence into a matrix built on all kinds of adjacent amino acids and applying the low rank approximation model to the obtained matrix to extract its feature representation. Then, drug molecules are encoded with fingerings which are a kind of binary fragment descriptors. Finally, two computational approaches are generally employed for studying the drug-target relations: the multiple classifiers learning algorithms and the machine learning algorithms based on Hadoop distributed platform. Our research will provide a new theoretical guidance to drug research and development.