RNA分子是一类重要的分子机器，然而人们对其空间结构的认识非常有限。鉴于通过实验手段测定其结构的困难和昂贵，本项目拟结合机器学习领域的新进展，从计算角度发展预测RNA三级结构的新方法，为人们更好的认识其生物功能提供新工具。 具体研究内容为： 1）结合RNA简化模型和隐马尔科夫模型，从RNA结构数据库中提取主链和侧链的空间信息，建立一个高效和准确的链生长方法； 2）发展基于多层前馈神经网络的 RNA三级结构打分函数，用于评估计算生成的RNA链和天然态结构的相似度； 3）发展基于深度学习的从序列预测contact map的方法，后期可拓展到预测其distance map，上述预测信息可和顺序模特卡罗方法相结合，构建RNA三级结构； 4）预测所得结构的进一步模拟优化。本研究取得的成果将用于搭建网络服务器，提供结构预测服务和软件下载，提高我国在相关领域的地位和影响力。

RNA is an important type of molecular machine, however, the knowledge of their tertiary structures is lacking. Since it is difficult and costly to determine their structures experimentally, we plan to develop new computational methods for RNA tertiary structure prediction, by combining the recent advances in the machine learning field. Our researches will provide new tools and facilitate the understanding the biological function of RNAs. Specifically, the contents include: 1) by combining an RNA coarse-grained model and the hidden Markov model, we will extract the spatial information from an RNA database and develop an efficient and accurate chain growth approach; 2) we will develop a scoring function for estimating the quality of RNA tertiary structures, based on training a multi-layer neural network; 3) we will construct a deep belief network to predict RNA contact map from its sequence, and this method can be extended to the distance map prediction in the future, the obtained information will be used to construct RNA tertiary structures based on a sequential Monte Carlo method; 4) the predicted structures will be subjected to further optimization by molecular dynamic simulations. The research results will be used to construct a web server, which will provide RNA tertiary structure prediction service to the public, to improve our influence on the related scientific fields.