scZAG: Integrating ZINB-Based Autoencoder with Adaptive Data Augmentation Graph Contrastive Learning for scRNA-seq...

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Single-cell RNA sequencing (scRNA-seq) is crucial for understanding cellular heterogeneity and disease.
• Clustering scRNA-seq data is challenging due to high dimensionality and sparsity.
• Existing methods often fail to capture the complex topological structures and continuity within scRNA-seq data.

Purpose of the Study:

• To develop a novel deep learning framework, scZAG, for improved cell clustering in scRNA-seq data.
• To address the limitations of conventional methods in handling data sparsity and complex topology.
• To enhance the interpretation of cellular states and disease mechanisms through accurate cell subpopulation identification.

Main Methods:

• A zero-inflated negative binomial (ZINB) model for denoising sparse and over-dispersed scRNA-seq data.
• An adaptive graph contrastive representation learning approach using APPNPGCN and graph contrastive learning.
• Clustering of low-dimensional latent representations using Kullback-Leibler divergence.

Main Results:

• scZAG effectively denoises scRNA-seq data using a ZINB model.
• The graph contrastive learning component captures cellular relationships, reflecting data continuity and topology.
• Superior clustering performance of scZAG was validated on 10 benchmark scRNA-seq datasets compared to state-of-the-art methods.

Conclusions:

• scZAG offers a robust deep learning framework for scRNA-seq data analysis.
• The method accurately identifies cell types by preserving complex topological structures and data continuity.
• scZAG represents a significant advancement in scRNA-seq data clustering and interpretation.