Machine learning enables identification of an alternative yeast galactose utilization pathway

Area of Science:

• Genomics
• Microbiology
• Computational Biology

Background:

• Understanding genotype-phenotype relationships is crucial in biology.
• A comprehensive dataset of yeast genomes, growth patterns, and environments was compiled for 1,154 strains across 1,049 species.

Purpose of the Study:

• To investigate how genomic differences drive phenotypic variations in yeast.
• To predict yeast growth on various carbon sources using machine learning.

Main Methods:

• A random forest algorithm was employed, trained on genomic, metabolic, and environmental data.
• Predictive models were developed to assess growth on different carbon sources, with a focus on galactose utilization.

Main Results:

• Machine learning accurately predicted yeast growth on carbon sources, with genomic and growth data being key predictors.
• A novel alternative pathway for galactose utilization was identified in several yeast species, distinct from the canonical GAL pathway.
• Growth on galactose was best predicted by combining genomic and growth data.

Conclusions:

• Machine learning is a powerful tool for exploring yeast genotype-phenotype evolution.
• The study uncovered novel biological insights into yeast metabolic pathways and evolutionary adaptations.