Applying Machine Learning to Classify the Origins of Gene Duplications

Area of Science:

• Genomics
• Evolutionary Biology
• Bioinformatics

Background:

• Whole-genome duplication (WGD) events are common in land plant evolution, leaving legacies in extant diploidized genomes.
• Genes originating from WGDs (paleologs) influence plant evolution through functional divergence, genetic diversity, and gene loss.
• Existing paleolog classification methods are limited by reliance on specific genomic features, accuracy, and computational demands.

Purpose of the Study:

• To develop a supervised machine learning approach for classifying paleologs from specific WGD events in diploidized plant genomes.
• To create a robust and efficient tool applicable across diverse paleopolyploidy histories, including overlapping WGDs.

Main Methods:

• Collected empirical genomic data (syntenic block sizes, etc.) from 27 plant species with varied paleopolyploidy histories.
• Developed simulations of syntenic blocks and paleologs to train a gradient boosted decision tree model.
• Implemented the Frackify (Fractionation Classify) tool for paleolog identification and classification.

Main Results:

• Frackify accurately identifies and classifies paleologs across a wide range of parameter spaces, including complex scenarios with multiple overlapping WGDs.
• Comparison with existing paleolog inference methods in six species demonstrated Frackify's high consistency and efficiency.
• The tool effectively combines multiple genomic features for rapid paleolog classification.

Conclusions:

• Frackify offers an accurate, efficient, and consistent method for classifying paleologs derived from whole-genome duplication events.
• This tool advances the study of paleologs and their impact on plant genome evolution.
• Frackify provides a valuable resource for researchers investigating ancient duplication events in plants.