By Wende Whitman
A photo of a rock wall.

Figure 1: Rock succession from which the sample originally was taken; West Australia, Pilbara region, part of the 3.48 billion years old Dresser Formation. The rock succession records an ancient coastal setting.

Featuring contributions from Old Dominion University sedimentologist Nora Noffke, Ph.D., a multidisciplinary team of researchers found that artificial intelligence (AI) can identify biochemical “fingerprints” in organic material that is more than three billion years old. Noffke joined researchers from the Carnegie Institution for Science, Harvard University, Howard University and others in a new study. 

Using an AI model trained on chemical signatures from over 400 modern, fossil, synthetic and meteoritic samples, the team achieved remarkably accurate classifications of biological versus non-biological material and even identified potential metabolic traits like photosynthesis in ancient rocks. The breakthrough offers a new way to probe the earliest evolution of life and strengthens tools that could one day detect life beyond Earth.  

A photo of the side of a rock wall.

Figure 2: Close-up of the above succession in Figure 1. The thin, repeated layers record individual flooding events of a coastal setting.

Nora Noffke, Ph.D., and the Samples Behind the Science 
Noffke, a leading expert on ancient microbial structures, supplied and vetted crucial ancient rock samples for the study. Her expertise in microbially induced sedimentary structures, the rock textures left behind by ancient microbes, and early microbial ecosystems helped ensure that the rocks fed into the model had reliable geological context and were known to be of biological origin. That input is vital in a study where the conclusions depend on accurate geological context. Without trusted samples, AI predictions cannot be meaningfully interpreted. 

“By combining rigorous geological context with new analytical tools, we’re able to extract biogenicity signals that were effectively invisible before,” Noffke said.  

A photo of the side of a rock.

Figure 3: Close-up of the above succession in Figure 2. Vertical section through a stack of ancient microbial mats here visible as the black laminae. The laminae covered ripple marks. Ripple marks are formed by bottom currents in sandy sediments.

Insights for Life’s Origins and the Search Beyond Earth  
If machine learning can detect trace biological “fingerprints” that survived billions of years within Earth’s most ancient rocks, the same approach could be used to analyze Martian samples or future samples brought back from nearby moons like Europa or Enceladus. The method could help resolve long-standing questions about when key evolutionary innovations, such as photosynthesis, first emerged and whether ancient organic molecules preserved in certain Archean rocks, formed more than 2.5 billion years ago, are genuinely biological in origin. Being able to evaluate such material with this level of sensitivity would shelp determine whether any detected organic signatures reflect biology rather than geology. 

A promising tool, not a replacement 
The authors are careful to operate within the current capabilities of AI.  

  • Some samples fall into a mid-range probability “gray zone,” where the model cannot make firm conclusions.
  • Larger, more balanced training sets are needed, especially more fossil animals and a broader range of abiotic reference materials.
  • The method complements traditional tools such as isotopic analyses and fossil morphology rather than replaces them. 
Fossil microbial mats.

Figure 4: Microscopic view of the above photo in Figure 3. The fossil microbial mats seen as dark laminae under the microscope. The little triangles are areas where gas bubbles pushed the mat layer upward.

Next steps 
The research team plans to refine their models, explore additional machine-learning architectures and expand the sample base. They also aim to test the AI approach on rocks collected from extreme, Mars-like desert environments on Earth. 

Learn more: 
To read the full announcement and additional commentary from the research team, visit the Carnegie Institution for Science’s .