A Bayesian factorization method to recover single-cell RNA sequencing data.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Single-cell RNA sequencing (scRNA-seq) enables high-throughput gene expression analysis.
• Sparse gene expression matrices with dropout events present a significant challenge in scRNA-seq data analysis.
• Accurate imputation is crucial for robust cell-to-cell variation studies and gene regulatory network reconstruction.

Purpose of the Study:

• To introduce Bfimpute, a Bayesian factorization imputation algorithm for scRNA-seq data.
• To address the matrix completion problem caused by dropout events in gene expression matrices.
• To improve the accuracy of gene expression imputation in scRNA-seq data.

Main Methods:

• Developed Bfimpute, a Bayesian factorization imputation algorithm.
• Reconstructed two latent gene and cell matrices for imputation.
• Applied imputation within cell groups, with or without cell type labels or bulk data.
• Integrated user-provided gene- or cell-related information.

Main Results:

• Bfimpute demonstrated superior accuracy compared to ten other scRNA-seq imputation methods.
• Performance was validated on both simulated and real scRNA-seq datasets.
• Evaluation metrics confirmed Bfimpute's effectiveness in recovering dropout events.
• Flexible integration of additional information further boosted imputation performance.

Conclusions:

• Bfimpute offers a robust and accurate solution for imputing dropout events in scRNA-seq data.
• The algorithm's flexibility allows for enhanced performance through integrated biological information.
• Bfimpute advances the analysis of scRNA-seq data, facilitating deeper biological discoveries.