A Random Forests Framework for Modeling Haplotypes as Mosaics of Reference Haplotypes.

Area of Science:

• Genomics and Bioinformatics
• Computational Biology
• Statistical Genetics

Background:

• Genomic data analyses like phasing, imputation, and ancestry inference rely on matching haplotype pairs to construct target haplotypes from reference panels.
• Current methods often use heuristic rules or hidden Markov models (HMMs), which may not fully capture complex genetic relationships.
• Local haplotype matching is a fundamental task across various genomic applications.

Purpose of the Study:

• To develop and evaluate an extremely randomized trees (extra-trees) framework for local haplotype matching.
• To compare the performance of the extra-trees framework against traditional hidden Markov models for haplotype reconstruction.
• To assess the utility of the extra-trees framework for whole-genome sequence imputation.

Main Methods:

• Developed a supervised classifier using extra-trees, a type of random forest, to learn optimal local haplotype matches from observed examples.
• Utilized 30 features related to linkage disequilibrium, linkage, and genealogical information, including segment length and relationship estimates.
• Employed repeated cross-validation to rank feature importance, identifying 'distance to the edge of a shared segment' as most critical.

Main Results:

• The extra-trees framework demonstrated superior efficiency compared to hidden Markov models in reconstructing target haplotypes as mosaics of reference haplotypes.
• When applied to whole-genome sequence imputation from 50k genotypes, the framework achieved average reliabilities comparable to or slightly better than IMPUTE2.
• Feature importance analysis highlighted key factors contributing to accurate local haplotype matching.

Conclusions:

• The extra-trees framework provides a novel and effective approach for automated local haplotype matching in genomic data.
• This method shows significant potential for improving tasks such as genotype imputation and haplotype reconstruction.
• The study lays the groundwork for future implementations and improvements in routine genomic analysis pipelines.