Analysis On Single-Cell RNA Sequence Data Based On Unsupervised Deep Learning

In recent years,the development of single-cell RNA sequencing(scRNA-seq)technology has allowed researchers to analyze and solve biological problems from the gene expression transcripts of single cells.Unsupervised deep learning can accurately and efficiently integrate and analyze massive single-cell sequencing data.During single-cell analysis,dimensionality reduction and clustering are essential steps in the analysis.Dimensionality reduction and clustering can remove redundant information in scRNA sequencing data and explore the type and number of cells.Due to the rapid development of neural network,scRNA sequencing data get rid of the constraints of traditional algorithms,such as time-consuming and low accuracy.The combination of unsupervised deep learning and scRNA sequencing data has brought single-cell sequencing research into a broader world.Based on unsupervised deep learning,this paper studies single-cell RNA sequencing data.The research contents are as follows:1.A new clustering method is proposed that uses wavelet noise reduction combined with a classical autoencoder network,which we call sc-WAE.First,suitable wavelets are selected to reconstruct scRNA sequencing data.A large number of nontrue zero values in the original data are reconstructed by wavelet into numerical values with practical meaning.Second,the reconstructed data is dimensionally reduced by the classical autoencoder network.On large datasets,unsupervised deep learning has unparalleled processing advantages.Autocoding network dimensionality reduction can quickly obtain low-dimensional and efficient scRNA sequencing data and improve the analysis efficiency of single-cell data.Finally,a variational Bayesian-Gaussian mixture model is used to cluster cells,and the clustering results are visualized.On the other hand,ground-truth cell labels are used to calculate and evaluate clustering performance.The ten ground-truth scRNA sequencing datasets are used to evaluate our proposed scWAE method.Compared with the other four clustering methods,it is proved that the clustering effect of sc-WAE method is better than the other four clustering methods.2.A clustering method based on double autoencoders combined with variational Bayesian-Gaussian mixture model is proposed,which we call sc-VBDAE.First,the scRNA-sequencing data are reconstructed by confronting the encoding and decoding processes of the autoencoding network.The data in the reconstructed single-cell gene expression matrix changed significantly,and the original zero value is reconstructed into non-zero value.Second,the classical autocoding network is used to reduce the dimensionality of the reconstructed scRNA sequencing data.The dimensionality reduction process obtains low-dimensional and efficient scRNA sequencing data,which in turn improves the analysis efficiency of scRNA sequencing data again.Then,scRNA-sequencing data are clustered using a variational Bayesian-Gaussian mixture model.Finally,the clustered scRNA-sequencing data are used for downstream analysis and visualization.The proposed method is sequentially run on ten representative singlecell datasets to obtain clustering results.According to the four clustering evaluation indicators,our method has a better clustering effect.At the same time,we run other five clustering algorithms and compare them with our proposed algorithm.The results show that our method outperforms other clustering methods overall.