药物-靶标相互作用的识别对于药物重定位、新药研发及药物毒副作用等领域都具有非常重要的意义。然而，受制于通量、精度和成本的影响，传统的生物实验方法通常难以完成大规模药物-靶标相互作用关系的识别。随着生物信息学的迅速蓬勃发展，计算机辅助的靶标识别预测方法越来越受到重视，发展快速、精确的靶标识别方法为创新药物先导结构的发现及筛选提供了一种有效手段。本项目拟在Spark平台下整合药物分子指纹结构信息和靶标蛋白氨基酸序列信息，构建基于深层次特征表示及双向长短时记忆模型对潜在药物-靶标相互作用进行预测研究。首先，研究药物分子及靶标蛋白氨基酸序列向量化编码新方法；然后，采用基于变分自动编码器神经网络模型对药物-靶标蛋白对的深层特征进行表征；最后，在Spark并行化计算平台下构建双向长短时记忆模型，以快速、准确地预测药物-靶标相互作用。本项目研究将为创新药物设计提供理论指导和高可信的数据支撑。

Identifying drug-target interactions is crucial in drug repositioning, innovative drug discovery and development, adverse side-effects of drugs and other fields. However, traditional biological experimental methods are difficult to predict large-scale drug-target interactions due to the limitations of throughput, accuracy and cost. With the rapid development of bioinformatics, the computational methods are more and more widely used in drug-target interaction predictions, which can provide an effective means for the discovery and screening of lead compounds. In this project, we explicitly attempt to propose a new method for predicting drug-target interactions based on the Spark parallel platform by combining the structure of drug molecular fingerprints and the information of amino acid sequences. Firstly, a new coding method is proposed for representing drug molecules and amino acid sequences of target proteins. Secondly, the variational autoencoder method is developed to characterize deep features of the essential attributes of data. Finally, the bi-directional long short-term memory model is developed to predict drug-target interactions on the Spark parallel platform. This project will provide highly reliable data support and theoretical basis for target drug design.