Predicting natural variation in the yeast phenotypic landscape with machine learning.

Area of Science:

• Genomics
• Systems Biology
• Computational Biology

Background:

• Organismal traits arise from complex genetic and environmental interactions, complicating prediction.
• Predicting phenotypes from genotypes is a significant challenge in biology.

Purpose of the Study:

• To explore phenotype predictions using machine learning (ML) models in Saccharomyces cerevisiae.
• To benchmark ML models for predicting 223 traits from genomic and gene expression data.

Main Methods:

• Utilized a machine learning pipeline with linear and non-linear models.
• Benchmarked gradient boosting machines as the top-performing model.
• Analyzed genotypes and gene expression data from 1011 yeast strains.

Main Results:

• Gradient boosting machines achieved the best prediction accuracy.
• Gene function disruption scores and gene presence/absence were significant predictors.
• Prediction accuracy varied by phenotype, with stress resistance more predictable than growth.

Conclusions:

• Machine learning effectively interprets the functional impact of genetic variants on phenotypes.
• The accessory genome plays a substantial role in controlling yeast phenotypes.
• Genomic information can be effectively transferred across related phenotypes.