scDBImpute: Dual-Branch Imputation for Single-Cell RNA-Seq Data Dropouts

Single-cell RNA sequencing (scRNA-seq) enables a comprehensive analysis of the expression patterns of individual cells in tissue with the resolution of a single cell. However, "dropouts" will lead to an excess of zeros in the scRNA-seq data due to technical constraints, which could hinder further analysis. Consequently, imputing the dropout values becomes particularly critical in assisting with the recovery of biological information. Herein, we propose a dual-branch imputation method for scRNA-seq data, which helps to impute the drops in scRNA-seq data. Unlike previous methods which assume a preconceived structure guiding to impute the dropouts, we believe that there are linear and non-linear associations that help structure the data, thus, both linear and non-linear pipelines are combined for dropout imputation. The evaluation and comprehensive downstream results on both simulated and real datasets show that our method outperforms the state-of-the-art methods for recovery of gene expression, cell clustering, differential expression analysis, and pseudo-time trajectory analysis tasks.