scGCL: an imputation method for scRNA-seq data based on graph contrastive learning.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Single-cell RNA sequencing (scRNA-seq) is crucial for understanding cellular heterogeneity and biological processes.
• High data sparsity and dropout events in scRNA-seq hinder accurate downstream analyses like cell clustering.
• Existing imputation methods fail to fully utilize cell-cell relationships and data topology.

Purpose of the Study:

• To develop an advanced imputation method for scRNA-seq data that overcomes limitations of existing approaches.
• To improve the accuracy of downstream analyses by effectively restoring gene expression levels.

Main Methods:

• Proposed scGCL, a method integrating graph contrastive learning and Zero-inflated Negative Binomial (ZINB) distribution for imputation.
• Utilized contrastive learning to capture global and local semantic information, enhancing node representations.
• Employed a ZINB-based autoencoder to model the global probability distribution and reconstruct scRNA-seq data.

Main Results:

• scGCL demonstrated superior performance compared to state-of-the-art methods across 14 scRNA-seq datasets.
• The method significantly improved clustering performance and gene imputation accuracy.
• scGCL enhanced the detection of specific gene expression patterns in Alzheimer's disease datasets.

Conclusions:

• scGCL offers a robust and effective solution for scRNA-seq data imputation.
• The integration of graph contrastive learning and ZINB distribution provides a powerful framework for handling sparse scRNA-seq data.
• This method facilitates more reliable downstream analyses and biological discoveries from scRNA-seq data.