An information-theoretic machine learning approach to expression QTL analysis.

Area of Science:

• Genomics
• Bioinformatics
• Systems Biology

Background:

• Expression Quantitative Trait Locus (eQTL) analysis links genotype to gene expression.
• Traditional eQTL methods often rely on linear regression models.
• Identifying both cis- and trans-eQTLs is crucial for understanding gene regulation.

Purpose of the Study:

• To develop a novel method for identifying eQTL associations using information theory and machine learning.
• To capture both linear and non-linear associations, as well as population heterogeneity.
• To compare the performance of the novel method against existing eQTL analysis tools.

Main Methods:

• Utilized Mutual Information (MI) to quantify the association between genetic markers and gene expression.
• Applied Maximum Relevance Minimum Redundancy (mRMR) and Incremental Feature Selection (IFS) for feature selection.
• Validated the method on apoE-deficient mice data and compared results with R/qtl and MatrixEQTL.

Main Results:

• The novel mRMR.eQTL method demonstrated higher concordance with other established methods compared to R/qtl in both female (67.9% vs 14.4%) and male (74.1% vs 18.2%) mice.
• Observed that cis-acting eQTLs were stronger but less numerous than trans-acting eQTLs in the studied mouse model.
• The method successfully identified associations, highlighting its potential for eQTL analysis.

Conclusions:

• The proposed information theory and machine learning-based method offers a new and effective approach for eQTL analysis.
• mRMR.eQTL provides a more robust and accurate identification of genetic associations with gene expression.
• The developed software is available, facilitating broader application in genetic research.