Inferring gene regulatory networks from single-cell gene expression data via deep multi-view contrastive learning

Area of Science:

• Genomics
• Computational Biology
• Systems Biology

Background:

• Gene regulatory networks (GRNs) are crucial for cellular function.
• Single-cell RNA sequencing (scRNA-seq) allows for high-resolution GRN inference.
• Existing methods often fail to leverage information from multiple data sources.

Purpose of the Study:

• To develop a novel model for inferring GRNs from multiple scRNA-seq datasets.
• To enhance GRN reconstruction by integrating diverse data sources.
• To address limitations of single-source network inference methods.

Main Methods:

• Proposed a multi-view contrastive learning (DeepMCL) model.
• Represented gene pairs as histogram images for deep Siamese convolutional neural network input.
• Incorporated an attention mechanism to integrate multi-source embeddings.

Main Results:

• DeepMCL demonstrated effectiveness in GRN inference using multiple data sources.
• Contrastive learning and attention mechanisms significantly improved performance.
• Validated on both synthetic and real-world scRNA-seq datasets.

Conclusions:

• DeepMCL offers a powerful approach for GRN inference by integrating multiple scRNA-seq data.
• The model's ability to leverage multi-view data enhances understanding of gene regulation.
• This method advances the field of single-cell-resolution GRN reconstruction.