Literature-based priors for gene regulatory networks

Area of Science:

• Bioinformatics
• Computational Biology
• Systems Biology

Background:

• Gene regulatory network (GRN) modeling benefits from prior knowledge integration.
• Existing methods face challenges in updating knowledge from rapidly growing scientific literature.
• Novelty lies in using literature-derived gene-pair association scores for Bayesian network learning.

Purpose of the Study:

• To present the first research on massive incorporation of prior knowledge from literature for Bayesian network learning of gene networks.
• To develop a method for transforming literature-based gene association scores into network prior probabilities.
• To evaluate the impact of literature-derived priors on GRN accuracy and biological interpretability.

Main Methods:

• Generated gene-pair association scores based on co-occurrence in scientific literature.
• Developed a method to convert these scores into network prior probabilities.
• Applied the method to learn gene subnetworks for yeast, E. coli, and human, investigating prior knowledge weighting.

Main Results:

• Literature-based priors significantly improved the number of true regulatory interactions.
• Enhanced accuracy in gene expression value prediction compared to networks learned solely from expression data.
• Learned networks demonstrated improved biological interpretation, with subnetworks aligning with known pathways.

Conclusions:

• Massive incorporation of literature-derived prior knowledge effectively enhances gene regulatory network modeling.
• The proposed method offers a scalable and interpretable approach to leverage vast scientific literature for GRN construction.
• This approach holds promise for advancing our understanding of gene regulation across different organisms.