DAE-TPGM: A deep autoencoder network based on a two-part-gamma model for analyzing single-cell RNA-seq data.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Single-cell RNA sequencing (scRNA-seq) is crucial for biomedical research but suffers from data sparsity and noise.
• Normalization is essential for scRNA-seq data, revealing semi-continuous patterns with right-skewed distributions.
• Existing dimensionality reduction and imputation methods often fail to capture these specific data characteristics.

Purpose of the Study:

• To develop an advanced method for joint dimensionality reduction and imputation of scRNA-seq data.
• To address the challenges posed by dropouts and noise in scRNA-seq datasets.
• To improve the accurate inference of gene expression patterns and cellular phenotypes.

Main Methods:

• Introduced a deep autoencoder network based on a two-part-gamma model (DAE-TPGM).
• The two-part-gamma model captures the statistical properties of semi-continuous normalized scRNA-seq data.
• The deep autoencoder leverages gene relationships for enhanced data imputation.

Main Results:

• DAE-TPGM demonstrated superior performance in capturing phenotypic information from peripheral blood mononuclear cells (PBMCs).
• The method effectively inferred continuous phenotype information for hematopoiesis in mice.
• DAE-TPGM significantly improved cellular phenotype recognition on two real datasets compared to MAGIC, SAVER, scImpute, and DCA.
• Analyses on synthetic data confirmed competitive advantages in uncovering gene expression patterns in time-course data.

Conclusions:

• DAE-TPGM offers a robust solution for scRNA-seq data analysis, excelling in dimensionality reduction and imputation.
• The model accurately captures complex gene expression patterns and enhances cellular phenotype identification.
• This approach provides a valuable tool for advancing biomedical studies utilizing scRNA-seq data.