Using stochastic causal trees to augment Bayesian networks for modeling eQTL datasets.

Area of Science:

• Genomics
• Systems Biology
• Bioinformatics

Background:

• Genotypic and genome-wide expression data enable complex phenotype modeling.
• Existing eQTL data modeling methods are limited in resolving causal relationships for distant transcripts.
• Bayesian networks augmented with genotypic data offer a powerful framework for causal inference.

Purpose of the Study:

• To present a probabilistic method for learning causal relationships between transcripts at all network levels.
• To enhance gene network reconstruction and expression quantitative trait loci (eQTL) mapping accuracy.
• To provide a prior for Bayesian network structure learning using causal information.

Main Methods:

• Developed a probabilistic method to infer causal relationships from genotypic and expression data.
• Utilized the inferred causal information as a prior for Bayesian network structure learning.
• Synthesized eQTL networks and corresponding data using established protocols.

Main Results:

• The proposed method significantly improves gene network reconstruction recall (20-90%) across various precision levels and dataset sizes.
• The learned networks enhance expression quantitative trait loci (eQTL) mapping, achieving over 10-fold increase in recall compared to traditional methods.
• Demonstrated improved performance over existing leading methods for eQTL network synthesis.

Conclusions:

• Integrating causal information as a prior for Bayesian network learning substantially boosts accuracy in gene network reconstruction and eQTL mapping.
• The method effectively establishes causal relationships for both proximal and distal transcripts.
• This approach advances the ability to model complex biological networks and genetic associations.