Function Prediction Of Long Non-coding RNA Based On ScRNA-seq Data

Long non-coding RNA(lncRNA),as an important type of non-coding RNA(ncRNA),has a wide range of biological functions in the life activities.For example,both human embryonic development and brain development are related to the spatio-temporal specific expression of lncRNA.Complex diseases,especially cancer,are closely associated with the mutations and disorders of lncRNAs.However,people's understanding of lncRNA is much less than that of protein-coding genes.The functional mechanism of most lncRNAs is still unknown.It is still an enormous challenge to predict the function of lncRNA and its potential relationship with cancer.In recent years,single-cell biology is one of the hottest research aspects.With the development of single-cell sequencing technology,single-cell RNA-seq(scRNA-seq)enable gene expression to be studied at an unprecedented resolution.At present,researches on lncRNA are mainly based on the tissue data,which will conceal the cell specific of lncRNA.The advent of scRNA-seq technology provide great support for analyzing the function of lncRNA at single cell level.However,due to the high dropout,low sequencing depth of scRNA-seq data,and the low expression of lncRNA compared with protein-coding genes,scRNA-seq data cannot be directly used for the analysis of lncRNA.Therefore,a method is needed to process scRNA-seq data so that scRNA-seq data can be used for the functional analysis and prediction of lncRNA.Based on scRNA-seq data,data mining,complex network models and statistics are used to study lncRNA function prediction methods and their associations with cancer.First,the cell uniform grouping method based on the resampling idea is proposed in the framework.Grouping based on cells can reduce the sparseness of scRNA-seq data and increase the expression level of lncRNA.Specifically,k-means algorithm is used to uniformly group cells according to the similarity between cells.Then the grouped cell are screened according to the similarity between cells,and the cell in the groups that meet the similarity standard are combined to obtain bigCells.Based on the bigCells,constructing gene co-expression networks and using a module mining algorithm based on greedy strategy to obtain gene modules.Finally,the obtained gene modules are analyzed from multiple perspectives such as functional enrichment,differential expression and network topology characteristics,which will provide valuable reference information for the function prediction of cell-typespecific lncRNA.Finally,the cell uniform grouping method and the lncRNA function prediction model are verified.First,the uniform cell grouping method is verified on the simulation data and real melanoma scRNA-seq data.The result shows that the method can effectively screen and merge a group of cells with higher similarity.The comparison of the data before and after merging proves that cell merging can significantly increase the expression level of lncRNA and the correlation between genes.Then,two gene co-expression networks is constructed based on two types of bigCell,and the module detection algorithm is performed.There are17 lncRNAs velidated to have associations with melanoma by databases and literatures.Through multiple statistical analysis of gene modules and the analysis of network topology characteristics,it is shown that the obtained modules are closely related to melanoma.In addition,the function of lncRNA in the modules is further verified combined with cell types which proves the effectiveness of this framework.Moreover,the differential expression analysis showed that more meaningful differentially expressed lncRNAs can be obtained by the cell uniform grouping method.In summary,the calculation method proposed in this thesis can effectively analyze and predict the function of lncRNA based on scRNA-seq data,and lay the foundation for further research on the function of lncRNA,reveal the role of lncRNA in cancer.