基于集成深度学习的植物lncRNA与miRNA互作关系预测研究

Non-coding RNAs serve important roles in the principal biological processes of living organisms, which has become the darling of the life sciences. They do not only work independently but also work together to regulate the biological characteristics through their interactions. The majority of studies have focused on defining the regulatory functions of microRNAs (miRNAs) and long non‑coding RNAs (lncRNAs). There are a number of relevant available prediction software for humans and animals. However, they cannot be applied directly due to the special nature of plants. The research on the no-coding RNA has been restricted in the field of agriculture, forestry and horticulture. We proposed some prediction software of pre-miRNA, mature miRNAs, miRNA targets and their functions. We found and verified some functions of miRNAs and lncRNAs. Based on the foregoing work, we will combine some intelligent optimization algorithms to predict RNA secondary structure and construct an effective and rich vector of the interaction features between lncRNAs and miRNAs; integrate the deep belief network based on their structural features and convolutional neural network based on interaction sequence, which is suitable for biological data analysis; predict the interactions between lncRNAs and miRNAs; verify some functions of key regulatory networks by biological experiments. The project will provide a tool for the research on the interactions between plant lncRNAs and miRNAs, and also open up new ways for the deep learning method to predict the interaction of non-coding RNA.

广泛存在于生物体内、具有强大功能的非编码RNA已成为生命科学领域的宠儿，它们既可单独又可通过彼此间的互作来调控生物的性状，尤其是miRNA和lncRNA。在二者的互作方面已有多款适于人类和动物的预测软件问世，但由于植物的特殊性不能直接套用，制约了农林园艺等诸多领域非编码RNA研究工作的开展。本项目在研制出植物miRNA前体及成熟体预测、靶基因预测、miRNA功能预测等软件，并利用它们发现、鉴定了很多植物miRNA及lncRNA的基础上，拟结合智能优化算法预测RNA二级结构，构造丰富有效的lncRNA与miRNA互作结构特征向量；构建适合生物学数据分析的基于深度信念网络和卷积神经网络的集成深度学习模型；预测lncRNA与miRNA的关系，结合生物学实验探明重要调控网络及其功能。以期为植物lncRNA与miRNA互作关系的研究提供工具，也为深度学习方法在非编码RNA相关预测上的应用开辟新途径。

微小RNA (miRNA)和长非编码RNA (lncRNA)在调控内源基因表达、转录翻译后修饰、表观遗传等方面发挥着重要的作用。尤其是miRNA和lncRNA还能够相互作用影响各种生命活动。本项目主要研究miRNA-lncRNA互作特征的表示方法、互作关系预测以及互作功能预测。首先，提出了基于信息强化的互作特征表示方法RNAI-FRID，分别从miRNA和lncRNA分子中提取多样化的基础特征，通过算术级方法构造复杂特征，融合多种特征排序方法实现自适应地特征选择。其次，提出了基于多级信息强化和贪婪模糊决策的集成深度学习模型PmliPEMG，采用融合复杂特征、多尺度卷积长短记忆网络和注意力机制分别从特征级、尺度级和模型级强化样本信息，基于贪婪模糊决策集成个体学习器的输出，得到最终的预测结果。最后，结合集成剪枝和生物技术构建调控网络进行功能预测，设计了双路并行集成剪枝模型DPEP以获得miRNA与lncRNA相互作用伪标签，结合生物学规则提炼标签，建立miRNA与lncRNA的关系，再结合专家知识和经验开发的生物信息学识别方法，建立miRNA与mRNA的关系，进而建立三种RNA之间的关系，接着结合分子生物学实验测定三种RNA的表达量，构建lncRNA-miRNA-mRNA调控网络，最后通过GO富集分析预测miRNA与lncRNA相互作用的功能。项目的顺利完成，为新的植物miRNA与lncRNA相互作用数据的获取，它们的互作机制和功能的探索以及针对于植物的相关数据库的建立提供有力的支持，进而对基因编辑、生物制药、食品加工等技术的进步和广泛应用具有重要意义。在该项目的支持下，发表学术论文32篇，其中27篇被SCI收录；申请发明专利3件（已授权1件）；培养了2名博士、7名硕士。