Tensor Decomposition-based Feature Extraction and Classification to Detect Natural Selection from Genomic Data.

Area of Science:

• Genomics
• Evolutionary Biology
• Bioinformatics

Background:

• Identifying adaptive events, like lactose digestion or viral evolution, is key to understanding traits.
• Traditional methods for detecting natural selection from genomic data have limitations in handling complex scenarios.
• Machine learning, particularly convolutional neural networks, shows promise but faces challenges in parameter estimation and feature localization.

Purpose of the Study:

• To introduce T-REx, a novel computational framework for detecting natural selection.
• To leverage tensor decomposition for feature extraction from haplotype data.
• To improve the accuracy and robustness of identifying adaptive evolutionary processes.

Main Methods:

• T-REx utilizes tensor decomposition to extract features from haplotype images across individuals.
• These extracted features are then fed into classical machine learning models for prediction.
• The method was tested on simulated data representing neutral evolution and selective sweep scenarios.

Main Results:

• T-REx demonstrated high power and accuracy in distinguishing selective sweeps from neutral evolution.
• The method proved robust against common technical challenges in genomic data analysis.
• T-REx offers straightforward visualization of feature importance, aiding interpretation.

Conclusions:

• T-REx is a powerful and effective tool for detecting adaptive processes in genomic data.
• The framework enhances the toolkit for evolutionary genomics research.
• This approach offers a promising alternative to existing methods for natural selection inference.