基于靶蛋白和短序列模体谱的植物lncRNA功能分析方法研究

Functional analysis of lncRNAs is a research hotspot and difficulty in current bioinformatics, which also plays an important role in deciphering molecular regulatory mechanisms of key traits in plants. Considering that lncRNAs perform their regulatory functions by targeting proteins through specific short sequences, our research focus on functional analysis methods of plant lncRNAs based on target proteins and short sequence motif profiles (SSMPs). Firstly, to identify targeting relationships of lncRNAs and proteins, a novel approach of predicting lncRNA-protein interaction is developed based on marginal fisher analysis and deep learning methods. Secondly, to discover the targeting relationships further from the perspectives of lncRNAs and proteins, two methods are put forward to investigate functional relationships of lncRNAs as well as proteins respectively; one detects functional relationships of lncRNAs based on SSMPs and the other measures functional similarities of proteins by fusing multiple similarities. Finally, based on the studies mentioned above, an efficient computational analysis model of plant lncRNA functions is proposed by incorporating Gene Ontology (GO) annotations of target proteins and semantic accessibility relationships of GO terms. This research will be of great theoretical significance for developing analysis methods of biological data in the intersection of information science and life science. Meanwhile, it also has potential value in molecular design of breeding for improving plant traits.

解析lncRNA的功能是当前生物信息学领域的研究热点和难点，同时这对于揭示植物关键性状的分子调控机制具有重要作用。由于lncRNA通过特异性短序列靶向作用于蛋白质行使其调控功能，本项目拟重点研究基于靶蛋白和短序列模体谱的植物lncRNA功能分析方法。首先，建立基于边际Fisher分析和深度学习的lncRNA-蛋白质相互作用预测模型，以准确识别lncRNA-蛋白质靶向关系。其次，提出基于短序列模体谱的lncRNA功能关联关系挖掘方法和基于多源相似性融合的蛋白质功能相似性度量方法，进而从lncRNA及靶蛋白层面深入挖掘lncRNA-蛋白质靶向关系。最后，以上述研究为基础，结合靶蛋白的基因本体（GO）注释以及GO语义可达关系，建立高效解析植物lncRNA功能的计算模型。本项目对研究信息科学与生命科学交叉的生物数据分析方法具有重要的理论意义，同时在植物性状改良育种的分子设计方面具有潜在的应用价值。

长链非编码RNA（lncRNA）是调控动植物生长发育的重要分子元件，与生物体内细胞的分化、增殖及凋亡等众多重要的生命进程密切相关。本项目以lncRNA-蛋白质之间的靶向调控关系为切入点，围绕lncRNA功能预测所涉及到的关键计算问题展开研究，取得的重要成果包括：（1）提出了一种基于边际Fisher和深度自动编码器的lncRNA-蛋白质相互作用。与同类方法相比，该方法构建的样本特征空间具有更好地表征能力和区分能力，从而能够更为精准地识别lncRNA-蛋白质之间的靶向调控关系。（2）开发了一种新的基于短序列模体谱的lncRNA功能相似性计算方法，并将其应用于乳腺癌和肺癌相关lncRNA预测。该方法突破了lncRNA功能相似性计算依赖序列之外的其他先验信息的局限性，具有优于同类方法的预测准确性，能够为识别疾病相关lncRNA提供有效的依据。（3）设计了一种基于相似性网络和靶向关系的多标签lncRNA功能预测方法，并预测HOTAIRM1和lncRNA GAS5的功能。该方法深度融合了lncRNA之间的拓扑关系特征及其自身的功能特征，进一步提升了lncRNA功能预测性能，预测结果与生物医学领域的研究结论一致性高。上述研究成果能够为生命科学领域探索lncRNA功能和调控机制提供线索指引，同时对信息科学领域发展多组学生物数据处理和分析方法具有重要的理论意义。