Comparison between instrumental variable and mediation-based methods for reconstructing causal gene networks in yeast

Area of Science:

• Systems Biology
• Genetics
• Bioinformatics

Background:

• Causal gene networks are crucial for understanding cellular information flow.
• Reconstructing these networks from omics data is challenging due to the correlation-causation problem.
• Genomics and transcriptomics data can orient causality using genomic variants.

Purpose of the Study:

• To compare instrumental variable, mediation, and coexpression methods for causal gene network inference.
• To evaluate method performance using a large yeast dataset and a known interaction database.
• To identify strengths and limitations of causal inference methods in network reconstruction.

Main Methods:

• Utilized the Findr software package for uniform implementation of network inference methods.
• Employed a dataset of 1012 segregants from a budding yeast cross.
• Validated results against the Yeastract database of known transcriptional interactions.

Main Results:

• Causal inference methods showed significant overlap with the ground-truth network; coexpression performed randomly.
• Mediation method performance saturated at large sample sizes due to residual correlations.
• Instrumental variable methods produced false positives from linked eQTLs but excelled at specific transcription factor targets.

Conclusions:

• Causal inference from genomics and transcriptomics is a powerful tool for gene network reconstruction.
• Further improvements require addressing residual correlations in mediation and genomic linkage in instrumental variable methods.
• Instrumental variable and mediation methods have complementary roles in identifying causal genes, especially in transcriptional hotspots.