scGCL

scGCL leverages graph contrastive learning and a Zero-inflated Negative Binomial (ZINB) autoencoder to impute dropout events in scRNA-seq data and improve downstream analyses such as cell clustering.

Key Features:

• Graph Contrastive Learning: Integrates graph-based techniques to capture global and local semantic information and enhances node representations via contrastive learning with positive sample selection.
• ZINB Autoencoder: Uses an autoencoder based on the Zero-inflated Negative Binomial (ZINB) distribution to model gene expression and reconstruct scRNA-seq data while accounting for dropout events.

Scientific Applications:

• Improved Clustering Performance: Demonstrated superior clustering results across 14 diverse scRNA-seq datasets compared to existing imputation methods.
• Gene Expression Enhancement: Enhances expression patterns of specific genes in disease-relevant data, including applications to Alzheimer's disease datasets.

Methodology:

Integrates graph-based representation learning and contrastive learning with positive sample selection to enhance node embeddings, and employs a ZINB-distribution-based autoencoder to reconstruct scRNA-seq data and estimate dropout values.