基于多模态深度学习的lncRNA-RNA相互作用模式分析及预测研究

During the post-genomic era, a growing number of lncRNA sequences are determined, how to identify their functions and action mechanisms is one of the most important issues. To execute various functions, lncRNAs could select a diversity of mechanisms by interacting with different biomolecules. The traditional biological experimental determination and validation on interactions are usually laborious and time-consuming. To timely and effectively discover these interactions when facing with the avalanche of new lncRNA sequences, the prediction methods based on computing are emerging. In this project, as lncRNA and RNA molecules are heterogeneous and complex from multiple sources, we will study from the following several aspects: firstly, the different similarity measures of different features of lncRNA and RNA molecules are presented. Secondly, sequence features, structure features, energy features, evolution features, and physical/biological chemical property features are computed and analyzed to find different motifs of lncRNA-RNA interactions that provide an interpretable theoretical basis for predicting lncRNA-RNA interactions. Furthermore, a novel method, multi-modal deep learning, is proposed for predicting lncRNA-RNA interactions based on Deep Learning and Multi-modal Learning. Finally, an effective method for predicting lncRNA-RNA interactions is presented, and case studies and genome-wide analysis are carried out using the built prediction models to provide new insights for computation and biology research.

在后基因组时代，越来越多的长链非编码RNA（lncRNA）序列被测定出来，如何确定lncRNA的功能和作用机制是最重要的问题之一。lncRNA只有与其它生物分子相互作用才能执行相应的功能。采用传统生化实验测定这些相互作用费时费力，在这种情况下，依靠计算进行相互作用预测的方法应运而生。本研究拟从lncRNA及与其相互作用的RNA分子多源、异构、复杂的特征入手，首先研究这些特征间的相似性度量问题；其次研究序列特征、结构特征、能量特征、进化特征、物理/生物化学属性特征等的提取计算分析方法，以寻找确定lncRNA与RNA分子相互作用在各种特征下的相似性模体，为相互作用提供可解释的理论依据；再结合多模态深度学习研究lncRNA与RNA相互作用的预测问题。最后整体形成有效的lncRNA与RNA相互作用预测方法，并基于构建的模型进行个案研究和全基因组分析，为计算和生物学研究提供新的认知。

在后基因组时代，如何确定lncRNA的功能和作用机制是最重要的问题之一。lncRNA只有与其它生物分子相互作用才能执行相应的功能。本研究从lncRNA及与其相互作用的RNA分子多源、异构、复杂的特征入手，首先研究了这些特征间的相似性度量问题；其次研究了序列特征、结构特征、能量特征、进化特征、物理/生物化学属性特征等的提取计算分析方法，以寻找确定lncRNA与RNA分子相互作用在各种特征下的相似性模体，为相互作用提供可解释的理论依据；再结合多模态深度学习研究了lncRNA与RNA相互作用的预测问题。最后整体形成了有效的lncRNA与RNA相互作用预测方法，并基于构建的模型进行了个案研究和全基因组分析，为计算和生物学研究提供了新的认知。