Deep Nonnegative Matrix Factorization Using a Variational Autoencoder With Application to Single-Cell RNA Sequencing...

Area of Science:

• Genomics
• Computational Biology
• Bioinformatics

Background:

• Single-cell RNA sequencing (scRNA-seq) is crucial for understanding cellular heterogeneity in biological phenomena.
• Variational autoencoders (VAEs) are increasingly used for scRNA-seq data analysis due to their scalability.
• Existing VAE models lack interpretability for biological pattern explanation.

Purpose of the Study:

• To develop an interpretable method for scRNA-seq data analysis.
• To decompose gene expression parameters into shared and cell-specific components.
• To improve biological interpretability and accuracy in scRNA-seq data analysis.

Main Methods:

• Proposed an interpretable nonnegative matrix factorization (NMF) method.
• Decomposed parameters into shared and cell-specific components.
• Applied a variational autoencoder (VAE) for nonlinear dimension reduction on cell-specific parameters and introduced log-regularization.

Main Results:

• Achieved effective nonlinear dimension reduction.
• Successfully estimated cell-type-specific gene expression.
• Demonstrated excellent performance in simulation and real data analyses.

Conclusions:

• The proposed interpretable NMF method enhances biological insights from scRNA-seq data.
• The model effectively balances nonlinear dimension reduction with biological interpretability.
• This approach offers a valuable tool for biomedical studies utilizing scRNA-seq.