Research On The Prediction Of Plant MiRNA-lncRNA Interactions

MicroRNAs(miRNAs)and long non-coding RNAs(lncRNAs)play important roles in regulating endogenous gene expression,post-translational modification and epigenetics.Mi RNAs and lncRNAs can also interact with each other to affect various life activities,which have attracted a lot of attention.Molecular biology experiments can identify miRNA-lncRNA interactions,but they are time-consuming and expensive.Computational methods can predict potential interactions,which filters out valuable data for molecular biology experiments to save a lot of time and costs.Existing related studies mainly focus on animal and pay little attention to plant.Since the differences in nucleic acid structure and sequence of animal and plant,the prediction methods for animal data are not reliable on plant data.The databases and reliable prediction methods of plant miRNA-lncRNA interactions are still lacked,and many of their functions are unknown.This thesis uses machine learning technologies to predict plant miRNA-lncRNA interactions and further predicts their functions.It provides support for biological research and building the plant databases,which is significant to progress and wide application of technologies in gene editing,biopharmaceuticals,food processing,and so on.Since existing plant miRNA-lncRNA interaction data and prediction methods are limited,the reliable dataset are constructed,and the prediction process based on supervised learning is investigated.The feature extraction,sequence encoding,training of traditional machine learning models,design and training of deep learning models,ensemble learning technology and implementation of prediction method based on combined model and fuzzy decision(Pmli Pred)are described.Experimental results show that Pmli Pred has better performance than existing prediction methods.Some new plant miRNA-lncRNA interactions are successfully identified from the predicted results of Pmli Pred by molecular biology experiment.This study gives the construction method of plant miRNA-lncRNA interaction dataset and takes Pmli Pred as an example to give the prediction process of plant miRNA-lncRNA interactions.This study is the initial effort to conduct the innovative research.Based on it,three important problems are studied respectively.(1)A representation method of interaction features based on information enhancement(RNAI-FRID)is proposed for the problem of lacking effective representation method of plant miRNA-lncRNA interaction features.Diverse base features are extracted from each of miRNA and lncRNA molecules to obtain more sample information.The complex features are constructed through an arithmetic-level method,which greatly reduces the dimension of feature vector while keeping the relationship between molecular features.Since the dimension reduction may cause sample information loss,an arithmetic mean strategy is adopted in complex feature construction to enhance the sample information further.Three feature ranking methods based on Random Forest,Extra Tree and Gradient Boosting are ingetrated for adaptive feature selection.Experimental results show that RNAI-FRID is an efficient representation method for plant miRNA-lncRNA interaction features,which is beneficial for training more powerful models.(2)An ensemble deep learning model based on multi-level information enhancement and greedy fuzzy decision(Pmli PEMG)is proposed for the problems of insufficient sample information in training and inability to utilize the characteristics of individual learners in ensemble of plant miRNA-lncRNA interaction prediction model.The fusion complex features,multi-scale Convolutional Long Short-Term Memory network and attention mechanism are used to enhance the sample informance at feature,scale and model levels respectively,and diverse individual learners are obtained.The outputs of the individual learners are intreated based on a greedy fuzzy decision to exert their characteristics,and the final predicted result is obtained.Experimental results verify the effects of multi-level information enhancement and greedy fuzzy decision,and show that Pmli PEMG has better performance than state-of-the-art methods on cross-species prediction of plant miRNA-lncRNA interactions.(3)The ensemble pruning and biological technologies are combined to construct the regulatory network for function prediction,which solves the problems of few known functions and high cost of function identification of plant miRNA-lncRNA interactions.A dual-path parallel ensemble pruning model(DPEP)is designed to obtain the pseudo labels of miRNA-lncRNA interactions.The biological rules are combined to refine the labels to build the relationship between miRNAs and lncRNAs.The bioinformatics identification method developed based on expert knowledge and experiences is combined to build the relationship between miRNAs and messenger RNAs(m RNAs).Then the relationships among three RNAs are built.The expression levels of these RNAs are measured by molecular biology experiment.The lncRNA-miRNA-m RNA regulatory network is constructed.The functions of miRNA-lncRNA interactions are predicted through GO enrichment analysis.Experimental results show that DPEP has better performance than state-of-the-art prediction methods.The regulatory network is successfully constructed through the combination of DPEP and biological technologies,and the effects of plant miRNA-lncRNA interactions on multiple life activities and biological functions are predicted.