scMEDAL for the interpretable analysis of single-cell transcriptomics data with batch effect visualization using a...

Area of Science:

• Genomics
• Computational Biology
• Bioinformatics

Background:

• Single-cell RNA sequencing (scRNA-seq) offers insights into cellular heterogeneity.
• Technical and biological batch effects confound scRNA-seq data analysis.
• Existing methods often discard batch effects instead of modeling them.

Purpose of the Study:

• To introduce scMEDAL (single-cell Mixed Effects Deep Autoencoder Learning), a novel framework for batch effect modeling in scRNA-seq data.
• To separately model batch-invariant and batch-specific effects.
• To improve accuracy and interpretability of scRNA-seq data analysis.

Main Methods:

• scMEDAL employs two complementary autoencoder networks: one for batch-invariant representation via adversarial learning, and a Bayesian autoencoder for batch-specific representation.
• The framework utilizes fixed- and random-effects autoencoders for retrospective analyses.
• Genomap projections enable prediction of cell expression across different batches.

Main Results:

• scMEDAL effectively suppresses batch effects while modeling batch-specific variations across diverse conditions (autism, leukemia, cardiovascular).
• The framework enhances data accuracy and interpretability.
• Retrospective analyses reveal the impact of biological and technical effects on cellular expression.

Conclusions:

• scMEDAL provides a valuable framework for deeper insights into data acquisition and cellular heterogeneity.
• Combining batch-agnostic and batch-specific latent spaces enables more accurate predictions of disease status, donor group, and cell type.
• scMEDAL advances scRNA-seq data analysis by modeling, rather than discarding, batch effects.