Cell Type Prediction Of Single-cell Transcrip Tome Sequencing Data Based On Subspace Alignment

With the rapid development of single-cell RNA sequencing(scRNA),the research of genomics has been focused on cells,which provides a new opportunity for the research of biological information with unprecedented resolution.Different types of cells constitute the heterogeneity of biological tissues,and the data of scRNA based on gene expression level enables people to understand the properties of individual cells and the heterogeneity of tissues.With the development of sequencing technology,how to utilize the existing information and data to preferably identify and analyze cell types and functions has become a key challenge.Here,a transfer learning algorithm for image recognition,subspace alignment domain adaptation(SA-DA),is applied to single-cell transcriptome sequencing data to achieve data dimension reduction and label migration by aligning the subspaces of source domain and target domain.Firstly,the source domain and the target domain are preprocessed,and the subspace coordinate system is constructed by principal component analysis.Secondly,the target aligned source coordinate system is obtained by subspace alignment,and the data is projected into the aligned subspace to obtain the low dimensional representation of the data.Finally,SVM classifier is established on the source domain data to migrate the category information from the source domain to the target domain for prediction.At the same time,batch effect correction was performed.Experiments on real data set,human peripheral blood mononuclear cells(PBMC)dataset,show that SA-DA algorithm can obtain a more friendly low dimensional space representation for classification task,has a certain correction effect on batch effect,and can be used for cell type prediction with high accuracy and stability.In addition,in the face of unbalanced data distribution and large differences in the number of different types of cells,SA-DA algorithm is superior to the comparison method in prediction accuracy and stability,and it also has good performance for the prediction of less cell types in the dataset.In summary,SA-DA algorithm is applied to single cell transcriptome sequencing data,which has significant effect in data dimension reduction,batch effect correction and cell type prediction.