Hierarchical boosting: a machine-learning framework to detect and classify hard selective sweeps in human populations

Area of Science:

• Population Genetics
• Genomics
• Machine Learning

Background:

• Detecting positive selection in genomic regions is crucial for understanding natural populations.
• Inconsistent results arise from various tests and populations, with few methods classifying selective event features like age or completeness.

Purpose of the Study:

• To develop a machine learning framework for enhanced detection and classification of positive selection events.
• To improve the consistency and detail of findings from genome-wide selection scans.

Main Methods:

• Developed a machine learning classification framework utilizing Hierarchical Boosting.
• Exploited combined abilities of selection tests to identify polymorphism features under hard sweep models.
• Controlled for population-specific demographic factors.

Main Results:

• Achieved high sensitivity in detecting hard selective sweeps.
• Provided insights into sweep completeness and age of onset.
• Generated a genome-wide classification map of hard selective sweeps for three human populations from the 1000 Genomes Project.
• Demonstrated higher sensitivity in African populations.

Conclusions:

• The framework overcomes classical selection vs. no-selection classification, explaining inconsistencies in existing methods.
• Offers a more nuanced understanding of positive selection events in human populations.