HDMC: a novel deep learning-based framework for removing batch effects in single-cell RNA-seq data.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Single-cell RNA sequencing (scRNA-seq) generates large gene expression datasets.
• Batch effects between datasets complicate integrated analysis.
• Existing methods struggle with accurate distribution matching and cell type alignment.

Purpose of the Study:

• To address challenges in batch effect removal for scRNA-seq data.
• To accurately reduce distribution differences between batches.
• To align samples across batches for robust cell type recovery.

Main Methods:

• Developed a hierarchical distribution-matching framework using a deep autoencoder.
• Employed adversarial training for global batch distribution matching.
• Utilized contrastive learning with maximum mean discrepancy for local distribution alignment and cluster recovery.

Main Results:

• The proposed method accurately matches global and local batch distributions.
• Contrastive learning successfully aligns similar cell clusters and separates dissimilar ones.
• Demonstrated superior performance over state-of-the-art methods on simulated and real scRNA-seq data.
• Showcased the ability to prevent overcorrection in batch effect removal.

Conclusions:

• The novel deep-learning approach effectively overcomes limitations in scRNA-seq batch effect correction.
• The method enables more accurate cell type identification across diverse datasets.
• Hierarchical distribution matching and contrastive learning offer a powerful solution for integrated scRNA-seq analysis.