Interpretable factor models of single-cell RNA-seq via variational autoencoders.

Area of Science:

• Computational biology
• Genomics
• Bioinformatics

Background:

• Single-cell RNA sequencing (scRNA-seq) enables the study of gene expression variability across individual cells.
• Understanding cell-type-specific expression patterns is crucial but requires scalable and interpretable statistical methods.

Purpose of the Study:

• To develop a statistical inference method for scRNA-seq data analysis that is both scalable and interpretable.
• To enable the identification of gene programs within massive scRNA-seq datasets.

Main Methods:

• Modification of a scalable variational autoencoder framework.
• Integration of factor models within the auto-encoding variational Bayes framework.

Main Results:

• The proposed approach offers interpretability without significant loss of accuracy.
• Successful identification of gene programs in large-scale scRNA-seq datasets.
• The method is domain-agnostic and applicable to various biological questions.

Conclusions:

• The developed factor model approach provides a powerful and accessible tool for scRNA-seq data analysis.
• This method enhances the interpretability of complex gene expression patterns in large biological datasets.
• The scVI package implements this factor model, facilitating its use in the research community.