Integrating single-cell RNA-seq datasets with substantial batch effects.

Area of Science:

• Computational biology
• Genomics
• Bioinformatics

Background:

• Single-cell RNA sequencing (scRNA-seq) data integration is crucial for robust analysis.
• Current methods face challenges harmonizing datasets across different species, organoids, primary tissues, and scRNA-seq protocols (e.g., single-cell vs. single-nuclei).
• Existing conditional variational autoencoder (cVAE) strategies for batch correction have limitations, such as ineffective regularization or removal of biological signals via adversarial learning.

Purpose of the Study:

• To develop an improved computational method for scRNA-seq data integration.
• To address limitations in current batch correction strategies for cVAE-based methods.
• To enhance the preservation and interpretation of biological signals in integrated scRNA-seq datasets.

Main Methods:

• Proposed sysVI, a cVAE-based integration method.
• Employed VampPrior and cycle-consistency constraints within the cVAE framework.
• Evaluated sysVI's performance in harmonizing diverse scRNA-seq datasets.

Main Results:

• sysVI successfully integrates scRNA-seq datasets across different systems (species, organoids, tissues).
• The method improves the biological signals within the integrated data.
• Enhanced biological signals facilitate downstream interpretation of cell states and conditions.

Conclusions:

• sysVI offers a robust solution for harmonizing heterogeneous scRNA-seq data.
• The method overcomes limitations of existing integration techniques.
• sysVI enhances the utility of scRNA-seq data for biological discovery.