Modeling causality for pairs of phenotypes in system genetics

Area of Science:

• Systems genetics
• Computational biology
• Bioinformatics

Background:

• Systems genetics aims to understand complex gene regulatory networks.
• Transcriptional regulation is key, but other factors like methylation also play a role.
• Unobserved regulatory mechanisms can reduce the accuracy of gene network reconstruction.

Purpose of the Study:

• To develop novel statistical tests for inferring causal direction between phenotypes.
• To address limitations of existing model selection criteria (AIC, BIC) by providing significance levels.
• To improve the accuracy of gene regulatory network reconstruction.

Main Methods:

• Extension of Vuong's selection tests for misspecified models.
• Development of tests for causal direction in the presence of unobserved variables.
• Evaluation via simulation studies and comparison with AIC, BIC, and other causality tests.
• Validation using yeast knockout experiment data.

Main Results:

• The proposed model selection tests demonstrate higher precision in causal inference.
• Significantly reduced false-positive rates compared to AIC, BIC, and a recent causality test.
• The tests provide a significance level, unlike AIC and BIC.
• Validation with yeast data confirmed improved accuracy.

Conclusions:

• The new statistical tests offer a more precise method for inferring gene regulatory causality.
• Reduced false positives are crucial for guiding expensive and time-consuming experimental validation.
• These tests enhance the reliability of systems genetics findings.