Prediction of Gene Regulatory Connections with Joint Single-Cell Foundation Models and Graph-Based Learning.

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Single-cell RNA sequencing (scRNA-seq) enables detailed gene regulatory network (GRN) inference but faces challenges due to data sparsity and noise.
• Supervised machine learning methods for GRN inference require extensive transcription factor-DNA binding data, which is often limited and expensive to obtain.

Purpose of the Study:

• To develop a robust framework for gene regulatory link prediction using scRNA-seq data.
• To leverage large-scale pre-trained single-cell foundation models (scFMs) and joint graph-based learning to overcome limitations of existing methods.

Main Methods:

• Proposed scRegNet, a novel framework integrating scFMs with joint graph-based learning for gene regulatory link prediction.
• Utilized vectorized gene-level representations to predict missing regulatory interactions.
• Evaluated performance on seven scRNA-seq benchmark datasets against nine baseline methods.

Main Results:

• scRegNet achieved state-of-the-art results, outperforming nine baseline methods on seven benchmark datasets.
• The proposed framework demonstrated superior robustness compared to baseline methods when trained on noisy data.

Conclusions:

• scRegNet provides a powerful and effective approach for accurate gene regulatory network inference from scRNA-seq data.
• The integration of scFMs and graph learning offers a promising direction for advancing GRN inference methodologies.