Refining computational inference of gene regulatory networks: integrating knockout data within a multi-task...

Area of Science:

• Computational Biology
• Genomics
• Systems Biology

Background:

• Gene regulatory networks (GRNs) are crucial for understanding cellular processes.
• Current computational methods for GRN inference struggle to integrate prior biological knowledge.
• Accurate GRN reconstruction is essential for deciphering complex biological systems.

Purpose of the Study:

• To develop a novel computational framework for enhanced gene regulatory network (GRN) reconstruction.
• To improve the utilization of existing topological information and prior knowledge in GRN inference.
• To pioneer the simulation of gene knockouts on bulk data within a GRN reconstruction context.

Main Methods:

• Proposed a graph neural network (GNN)-based Multi-Task Learning framework (MTLGRN).
• Encoded gene promoter sequences and biological features, concatenating representations.
• Constructed a multi-task learning framework including GRN reconstruction, gene knockout prediction, and gene expression matrix reconstruction.

Main Results:

• MTLGRN demonstrated superior performance over state-of-the-art baselines in GRN reconstruction.
• The framework effectively leveraged biological knowledge, including promoter characteristics and gene knockout information.
• MTLGRN successfully integrated gene knockout simulation with bulk data analysis for GRN inference.

Conclusions:

• MTLGRN offers a powerful approach for accurate GRN reconstruction by integrating diverse biological data.
• The framework enhances the comprehensive understanding of gene regulatory relationships.
• MTLGRN represents a significant advancement in computational approaches for systems biology research.