Boosting alternating decision trees modeling of disease trait information

Area of Science:

• Genetics
• Machine Learning
• Statistical Genetics

Background:

• Quantitative trait linkage (QTL) analysis is crucial for identifying genes influencing complex traits.
• Machine learning methods offer novel approaches to disease status modeling and genetic analysis.
• Simulated datasets, like the Genetic Analysis Workshop, are valuable for testing analytical methods.

Purpose of the Study:

• To apply Alternating Decision Trees (ADTrees) for improved quantitative trait linkage (QTL) analysis.
• To develop a novel quantitative trait measure using ADTrees-based prediction scores.
• To evaluate the effectiveness of ADTrees in detecting linkage evidence for complex diseases.

Main Methods:

• Applied ADTrees using 12 binary phenotypes and sex to predict Kofendrerd Personality Disorder status.
• Generated average prediction scores as a new quantitative trait for genome-wide QTL analysis.
• Utilized four ADTrees modeling strategies and expectation-maximization Haseman-Elston QTL analysis.
• Performed 10 runs of 10-fold cross-validation for robust score computation.

Main Results:

• Detected significant linkage evidence (p < 0.01) on chromosomes 1, 3, 5, and 9.
• Identified relevant phenotypes across four populations, with minor exceptions.
• ADTrees models revealed subgroup structures consistent with latent traits.

Conclusions:

• ADTrees provide a potentially more accurate disease status representation for genetic analysis.
• The method facilitates enhanced detection of linkage evidence in complex genetic studies.
• ADTrees show promise for integrating multiple phenotypes and clinical data in genetic association studies.