A Method Of Normalization And Transformation For Single Cell RNA Sequencing Data

Single cell RNA sequencing(scRNA-seq)technology is a study that applies RNA-seq technology to individual cells transcriptome.Since the amount of RNA from a single cell is very limited,scRNA-seq relies heavily on RNA amplification.Such heavy RNA amplification can cause random dropout events,which are mainly reflected in a large number of zero counts in the expression matrix.At the same time,the scRNA-seq data needs to be consistent with the basic assumptions of downstream analysis methods.Therefore,more specialized normalization and transformation methods are required for scRNA-seq data analysis.This paper summarizes normalization and transformation methods for the current mainstream scRNA-seq data,and further proposes a normalization and transformation method based on the same distribution hypothesis of stably expressed genes.This method provides a novel tool for selecting stably expressed genes.It can remove the technical noise in the stably expressed genes and make the normalized and transformed stably expressed genes closer to the same distribution.In addition,the method ensures cell biological heterogeneity in downstream analysis and improves accuracy in downstream analysis such as dimensionality reduction,clustering,and differentially expressed genes analysis.This paper adopts the auto-encoders in deep learning as a technical tool for normalization and transformation,and puts forward some feasible ideas for the further improvement of the auto-encoders performance.In this paper,five public scRNA-seq datasets are selected as the processing datasets in the empirical analysis.After normalization and transformation,it shows the corresponding results in four aspects,including the convergence and stability of the methods,the selection of stably expressed genes,the transformation results of stably expressed genes,and cell heterogeneity analysis.The results show that normalization and transformation method proposed in this paper for the scRNA-seq data is useful and has certain application value.