MAGIC: A Multimodal Adaptive GRN Inference Constructor.

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Gene regulatory networks (GRNs) are crucial for understanding cellular processes and disease.
• Inferring GRNs from single-cell RNA sequencing (scRNA-seq) data is challenging due to data sparsity and sparse network connectivity.
• Existing methods struggle to fully leverage multimodal biological data for accurate GRN inference.

Purpose of the Study:

• To develop a robust method for inferring gene regulatory networks (GRNs) from scRNA-seq data.
• To address the challenges of data sparsity and sparse connectivity in GRN inference.
• To integrate multimodal data sources for enhanced GRN construction.

Main Methods:

• Proposed the Multimodal Adaptive GRN Inference Constructor (MAGIC) method.
• Integrated gene expression, sequence, and semantic features.
• Employed a shared graph attention weight alignment and a Knowledge-Aware Multimodal Fusion Module.
• Constructed a dual-topology network by integrating consensus similarity and known GRNs.

Main Results:

• MAGIC achieved an average AUROC of 0.839 across seven scRNA-seq datasets.
• Outperformed state-of-the-art GRN inference models.
• Demonstrated robustness on spatial transcriptomic data for bladder and breast cancer.
• Successfully uncovered potential associations between transcription factors and target genes.

Conclusions:

• MAGIC effectively integrates multimodal data to improve GRN inference accuracy.
• The method addresses key challenges in scRNA-seq data analysis for GRN construction.
• MAGIC shows promise for identifying regulatory relationships in complex biological systems and diseases.