scHSC: enhancing single-cell RNA-seq clustering via hard sample contrastive learning.

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Single-cell RNA sequencing (scRNA-seq) offers high-resolution transcriptome insights.
• Challenges include large datasets and high dropout rates, impacting clustering and cell annotation.
• Existing methods struggle with scRNA-seq data complexity.

Purpose of the Study:

• To develop an advanced deep learning method for accurate scRNA-seq data clustering.
• To address challenges posed by data size and dropout events.
• To enhance cell type annotation through improved clustering.

Main Methods:

• Proposed scHSC, a deep learning approach utilizing hard sample mining via contrastive learning.
• Integrated gene expression and topological structure information.
• Employed an adaptive weighting strategy combining contrastive learning with a ZINB model.

Main Results:

• scHSC demonstrated significant superiority in clustering performance across 18 real scRNA-seq datasets.
• Outperformed existing deep learning-based clustering methods.
• Effective in handling challenges of scRNA-seq data.

Conclusions:

• scHSC offers a robust and accurate solution for scRNA-seq data clustering.
• The method effectively integrates diverse data features for improved biological insights.
• Provides a valuable tool for single-cell data analysis.