scDecorr: feature decorrelation based representation learning enables self-supervised alignment of multiple...

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Single-cell RNA sequencing (scRNA-seq) reveals cellular heterogeneity but faces computational challenges like sparsity, high variability, and batch effects.
• Integrating scRNA-seq data from diverse sources is crucial for comprehensive biological insights but is hindered by distributional differences between datasets.

Purpose of the Study:

• To develop a novel framework, scDecorr, for robust representation learning and data integration in scRNA-seq analysis.
• To address the computational challenges in scRNA-seq data analysis, including batch effects and data integration from diverse sources.

Main Methods:

• scDecorr employs feature decorrelation-based self-supervised learning (SSL) to generate low-dimensional cell representations without cell-type labels.
• The framework incorporates unsupervised domain adaptation, utilizing domain-specific batch normalization to achieve domain-invariant representations for effective data integration.
• Maximizing similarity among distorted embeddings while decorrelating components allows scDecorr to capture biological signatures and mitigate technical noise.

Main Results:

• scDecorr successfully integrates scRNA-seq batches from diverse sources without compromising biological variance, leading to improved clustering.
• The generated cell representations demonstrate robustness in label transfer tasks, enabling effective cell-type label transfer between reference and query datasets.
• The framework provides efficient analysis and integration of large, complex scRNA-seq datasets.

Conclusions:

• scDecorr offers a powerful solution for robust representation learning and data integration in scRNA-seq analysis.
• The method effectively handles batch effects and distributional differences, facilitating deeper understanding of cellular processes and disease mechanisms.
• scDecorr advances the field by providing a tool for efficient analysis of complex single-cell data.