Machine learning predicts biological system evolution by gene gains and losses

Prediction of evolution is a fundamental goal of biology with a potential impact on strategic pathogen control and genome engineering. While predictability of short-term and sequence-level evolution has been investigated, that of long-term and system-level evolution has not been systematically examined. Here, we show that evolution of metabolic systems by gene gains and losses is generally predictable by applying ancestral gene content reconstruction and machine learning techniques to ~3000 bacterial genomes. Our framework, Evodictor, successfully predicted gene gain and loss events at the branches of the reference phylogenetic tree, suggesting universally shared evolutionary pressures and constraints on metabolic systems. I herein present the mathematical model of Evodictor and our findings on evolutionary rules from physiological and ecological aspects. I will further discuss potential versatility of Evodictor approach to analyze various diversification processes along branching lineage trees, not only evolution, but also developmental processes.