Robust self-supervised learning strategy to tackle the inherent sparsity in single-cell RNA-seq data.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Single-cell RNA sequencing (scRNA-seq) is crucial for understanding cellular heterogeneity.
• Data sparsity in scRNA-seq limits cell type annotation and transcriptomic analysis accuracy.
• Existing methods struggle to fully overcome information loss due to sparsity.

Purpose of the Study:

• To introduce scRobust, a robust self-supervised learning strategy designed to mitigate scRNA-seq data sparsity.
• To enhance the accuracy of cell type annotation and transcriptomic analysis.
• To generate high-quality cell embeddings for diverse biological applications.

Main Methods:

• scRobust utilizes a Transformer architecture.
• It incorporates a novel self-supervised learning approach combining contrastive learning and gene expression prediction.
• Effectiveness was validated across nine benchmarks, dropout scenarios, and combined datasets.

Main Results:

• scRobust significantly outperformed existing methods in cell-type annotation tasks.
• Generated cell embeddings captured complex clustering information, including cell types and HbA1c levels.
• Embeddings facilitated the identification of marker genes associated with drug tolerance stages.
• The method also produced high-quality embeddings for single-cell Assay for Transposase-Accessible Chromatin sequencing (scATAC-seq) data.

Conclusions:

• scRobust offers a powerful solution for addressing sparsity in scRNA-seq data.
• The method demonstrates significant improvements in cell type annotation and biological interpretation.
• Its robust cell embedding generation extends utility to other single-cell modalities like scATAC-seq.