Imputation for Single-cell RNA-seq Data with Non-negative Matrix Factorization and Transfer Learning.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Single-cell RNA sequencing (scRNA-seq) offers high resolution for studying cellular heterogeneity and transcriptome dynamics.
• A significant challenge in scRNA-seq data is the high frequency of zero counts (dropout events), which impedes accurate downstream analysis.

Purpose of the Study:

• To develop an advanced imputation method for scRNA-seq data that overcomes the limitations of excessive zero counts.
• To enhance the accuracy of gene expression estimation and preserve biological relationships within the data.

Main Methods:

• The study proposes a novel method, Non-negative Matrix Factorization and Transfer Learning (NMFTL).
• NMFTL integrates non-negative matrix factorization with transfer learning, leveraging external datasets to improve imputation.
• Graph-regularized terms are incorporated to maintain the data's intrinsic geometric structure and enhance imputation accuracy.

Main Results:

• NMFTL demonstrates superior performance compared to existing matrix-factorization-based imputation methods in recovering missing gene expression values (dropout entries).
• The method effectively preserves both gene-to-gene and cell-to-cell relationships within the scRNA-seq data.
• NMFTL shows strong performance in downstream analyses, including cell clustering.

Conclusions:

• NMFTL is a robust and effective method for imputing scRNA-seq data, significantly improving upon existing techniques.
• The integration of transfer learning and graph regularization enhances the accuracy and biological relevance of imputed gene expression data.
• The NMFTL method has been implemented in R scripts and is publicly available for researchers.