EM-random forest and new measures of variable importance for multi-locus quantitative trait linkage analysis

Area of Science:

• Genetics
• Bioinformatics
• Statistical genetics

Background:

• Standard random forest (RF) variable importance (VI) is suboptimal for linkage analysis due to marker correlation.
• Identical by descent (IBD) distribution and marker ambiguity require specialized methods for quantitative trait linkage analysis.

Purpose of the Study:

• To develop an EM-random forest (EMRF) method for quantitative trait linkage analysis.
• To introduce novel VI indices that leverage marker correlations within IBD data.
• To improve marker selection accuracy in linkage analysis.

Main Methods:

• Developed an EM-random forest (EMRF) model incorporating Haseman-Elston quantitative trait linkage analysis.
• Defined new VI indices accounting for marker ambiguity and posterior IBD distribution.
• Weighted sib-pairs based on posterior IBD distribution.

Main Results:

• EMRF with new VI indices outperformed the original RF VI index in simulations.
• EMRF performed comparably or better than EM-Haseman-Elston regression LOD scores across various genetic models.
• Tree size and marker subset size are critical parameters for RF performance.

Conclusions:

• The novel EMRF method enhances quantitative trait linkage analysis by addressing marker correlation and ambiguity.
• The new VI indices provide a more accurate assessment of marker importance in linkage studies.
• EMRF offers a robust alternative for genetic linkage analysis, with available source code.