Bold claim: tracing bacteria’s long, hidden history is possible—and it reframes how we understand life on Earth. And this is where AI changes the game. Here’s how scientists used machine learning to map the evolution of bacteria and reveal surprising twists you won’t want to miss.
Bacteria are the planet’s most diverse and ancient life forms. They consist of single cells, lack bones, and don’t leave obvious fossils like larger animals. That makes it tough to build a precise timeline of their early evolution. Yet by combining artificial intelligence with geological and biological data, researchers have filled in many gaps. A recent study published in Science shows that some bacteria developed the ability to use oxygen long before Earth’s atmosphere became saturated with it about 2.4 billion years ago.
A turning point in Earth’s history
Earth formed about 4.5 billion years ago after a colossal impact that likely erased any existing life, leaving behind the first ancestors of all living beings: single-celled microbes. For much of life’s early history, these microbes dominated the planet. The famous aphorism that all biology must be understood through evolution underscores why this early period matters. But tracing the maneuverings of ancient life requires more than fossils; it requires clever inference across time.
DNA comparisons among today’s diverse life forms reveal relationships—humans are more closely related to certain fungi than to some plants, for instance—and they also help map bacterial relationships. However, DNA alone can’t tell us when specific evolutionary events occurred. A lineage’s split might have produced many descendants, but the exact age of that split remains uncertain.
Geology provides a crucial timestamp
Geology identifies a major Earth-wide event about 2.4 billion years ago: the Great Oxidation Event. Cyanobacteria began using photosynthesis, releasing oxygen as a byproduct. Over millions of years, atmospheric oxygen rose. Before this, most life faced oxygen as a poisonous challenge. Some bacteria adapted to oxygen, while others retreated to sheltered environments. This shift had profound consequences for evolution and the diversification of life.
Building a bacterial timeline with AI
To anchor bacterial evolution in time, researchers trained an artificial intelligence model to predict whether a bacterium uses oxygen based on its genes. Today’s oxygen-using bacteria, such as cyanobacteria in the oceans, provide clues about ancient metabolism. Many gut bacteria, in contrast, thrive without oxygen. The metabolism-driven genomic changes leave telltale signals in the DNA, making this a natural target for machine learning.
With the model trained, the team extended the approach to ancient lineages: by analyzing gene content, they inferred which ancestral bacteria were likely oxygen-users. Modern methods also allow estimating which genes ancestors carried, not just the relationships among present-day species. This combination of data sources enables a dated reconstruction of the bacterial tree of life.
A surprising twist emerges
Using the Great Oxidation Event as a real-world calibration point yielded a detailed bacterial timeline. The study, integrating geology, paleontology, phylogenetics, and machine learning, refined the timing of bacterial evolution in meaningful ways. Surprisingly, some lineages capable of using oxygen appear to have existed about 900 million years before the Great Oxidation Event, suggesting early bacteria evolved oxygen tolerance even when atmospheric oxygen was scarce.
Even more striking is the finding that cyanobacteria may have evolved the ability to use oxygen before they developed photosynthesis themselves, challenging simple assumptions about the sequence of metabolic innovations.
What this means for our understanding
This work demonstrates how life’s capabilities evolved in response to Earth’s changing environment, and it shows how AI can complement geological and biological evidence to reconstruct deep time. The integrated approach offers a richer, more nuanced view of bacterial history than any single discipline could provide.
About the authors
Ben Woodcroft is Associate Professor of Microbial Informatics at Queensland University of Technology. Adrián A. Davín is a Postdoctoral Fellow in the Department of Biology at the Swiss Federal Institute of Technology Zurich. This article first appeared in The Conversation.
