ScAEQN: A Batch Correction Joint Dimension Reduction Method on scRNA-seq Data

Abstract

As single cell sequencing technique continues to advance, the size of scRNA-seq dataset has been enlarging, generating batch effects that affect downstream analysis, such as clustering analysis and differential expression gene (DEG) analysis. In this context, we present a novel batch integration joint dimensionality reduction method titled scAEQN. It adopts QuantNorm to calculate coefficient matrix and constructs an autoencoder to estimate the matrix of coefficient, ultimately obtaining a low-dimensional representation and reconstruction of the data. scAEQN is compared to different batch correction methods on a simulated dataset and six real single-cell RNA datasets. The results suggest that scAEQN is superior to batch correction methods under comparison in downstream analysis. scAEQN effectively eliminates batches and strongly reserves clustering pattern of cells, providing solid back-up for downstream analyses. scAEQN enhances the capability of clustering and selects more representative and stable DEGs in differential expression gene analysis. The source code and supplementary information of scAEQN are provided on website https://github.com/SiningSong/scAEQN.