Attention-Guided Probabilistic Diffusion Model for Generating Cell-Type-Specific Gene Regulatory Networks from Gene...

Area of Science:

• Computational Biology
• Genomics
• Systems Biology

Background:

• Gene regulatory networks (GRNs) control cellular identity and function via gene transcription.
• Single-cell technologies enable dissection of regulatory mechanisms in specific cellular states.
• Existing computational methods often infer local regulatory interactions independently, limiting global perspective.

Purpose of the Study:

• To propose Planet, a deep learning framework for constructing cell-specific GRNs from single-cell RNA sequencing (scRNA-seq) data.
• To provide a systems-level view of how protein regulators orchestrate transcriptional programs.
• To enhance structural consistency of GRNs at a global level by jointly optimizing local network structures and gene expression profiles.

Main Methods:

• Developed Planet, a deep learning framework based on diffusion models.
• Decomposed GRN generation into Markovian evolution steps.
• Introduced a Triple Hybrid-Attention Transformer to capture long-range regulatory dependencies across diffusion time-steps.
• Employed a fast-sampling strategy for accelerated inference.

Main Results:

• Planet achieves competitive performance against state-of-the-art methods on multiple scRNA-seq datasets.
• Demonstrated improved global consistency in reconstructed GRNs.
• Successfully reconstructed a cell-type-specific GRN for mouse-lung Cd8+Gzmk+ T cells, identifying known and novel regulators.
• Delineated dynamic immunoregulatory changes associated with aging.

Conclusions:

• Planet provides a practical framework for constructing cell-specific GRNs with enhanced global consistency.
• Offers a complementary perspective to existing methods for GRN inference.
• Provides new insights into regulatory dynamics in health and disease, particularly in aging immune cells.