Leveraging data-driven self-consistency for high-fidelity gene expression recovery.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Single-cell RNA sequencing (scRNA-seq) is crucial for cell state determination and subtype classification.
• Current scRNA-seq analysis omits low-expression genes, causing inaccurate gene counts and hindering downstream analyses.
• Imputing missing expression values in large scRNA-seq datasets is computationally challenging.

Purpose of the Study:

• To introduce a data-driven framework, the self-consistent expression recovery machine (SERM), for imputing missing gene expression values in scRNA-seq data.
• To address the challenge of recovering omitted expression values in large scRNA-seq datasets.

Main Methods:

• SERM utilizes a neural network to learn data distributions from noisy scRNA-seq data.
• The framework imposes self-consistency on the expression matrix to recover missing values.
• Ensures similar expression level distributions across the matrix for accurate imputation.

Main Results:

• SERM significantly improves the accuracy of gene imputation compared to existing methods.
• Demonstrates orders of magnitude enhancement in computational efficiency.
• Facilitates more accurate gene counts and improved downstream analysis for scRNA-seq data.

Conclusions:

• SERM offers an effective solution for imputing missing gene expression data in scRNA-seq.
• The method enhances both accuracy and computational performance in gene expression recovery.
• Improves the reliability of cell subtype classification and other downstream analyses.