scGANCL: Bidirectional Generative Adversarial Network for Imputing scRNA-Seq Data With Contrastive Learning.

Area of Science:

• Genomics
• Computational Biology
• Bioinformatics

Background:

• Single-cell RNA sequencing (scRNA-seq) provides high-resolution cellular insights.
• Technical noise causes dropout events, complicating scRNA-seq data analysis.
• Existing deep learning imputation methods struggle with rare cell type identification.

Purpose of the Study:

• To develop an advanced imputation method for scRNA-seq data.
• To improve the accuracy of gene expression profile reconstruction.
• To enhance the identification of rare cell populations.

Main Methods:

• A novel self-supervised deep learning model, scGANCL, was developed.
• scGANCL integrates bidirectional generative adversarial networks (BiGAN) with contrastive learning (CL).
• Contrastive learning enhances cell representation by minimizing data distribution discrepancies.

Main Results:

• scGANCL demonstrated superior imputation performance across ten simulated and seven real scRNA-seq datasets.
• The model consistently outperformed seven state-of-the-art imputation methods.
• Ablation studies confirmed the effectiveness of individual model components.

Conclusions:

• scGANCL offers a robust solution for scRNA-seq data imputation, addressing dropout events.
• The model significantly improves downstream analysis, particularly for rare cell type detection.
• This approach advances the reliable interpretation of single-cell gene expression data.