Machine learning enables prediction of metabolic system evolution in bacteria

Area of Science:

• Evolutionary biology
• Genomics
• Systems biology

Background:

• Predicting long-term, system-level evolution remains a challenge.
• Previous studies focused on short-term, sequence-level evolution.
• Understanding evolutionary predictability has implications for pathogen control and synthetic biology.

Approach:

• Developed Evodictor, a framework combining ancestral gene content reconstruction and machine learning.
• Applied the framework to ~3000 bacterial genomes to analyze metabolic system evolution.
• Investigated pathway architectures and analyzed metagenomic datasets to confirm findings.

Key Points:

• Gene content evolution in bacterial metabolic systems is generally predictable.
• Evodictor accurately predicted gene gain and loss across phylogenetic branches.
• Evolutionary patterns are linked to physiological and ecological factors, including functional dependencies and habitat changes.
• Intraspecies gene content variation also proved predictable within the framework.

Conclusions:

• Evolutionary pressures and constraints on metabolic systems are widely shared across bacteria.
• The predictability of evolution extends to ongoing changes within extant species.
• This work provides a powerful tool for predicting evolutionary trajectories in biological systems.