Chemistry paired with AI has been employed to identify chemical indicators of life in ancient rocks dating back 3.3 billion years. This method could allow researchers to uncover previously unreachable biomolecular insights from rocks over 1.7 billion years old, aiding in deciphering the enigmas surrounding the origins of life and the development of its biochemical mechanisms, including photosynthesis, as well as assist in the quest for extraterrestrial life.
Fossils from ancient microorganisms and isotopic carbon signatures suggest that the earliest life forms on Earth originated roughly 3.45 billion years ago. However, there is limited biochemical proof of life preserved in ancient rocks that have endured billions of years of geological transformation. The earliest clear evidence of complex biomolecules like lipids and porphyrins – which play roles in early chemistry compartmentalization and metabolic pathways, respectively – dates back to around 1.7 billion years ago, creating a substantial void in the biochemical record covering half of the known history of life.
Now, an international group has attempted to bridge this void by utilizing analytical chemistry and machine learning to extract biosignatures from rocks much older than 1.7 billion years. “Unlike prior studies, we are not targeting specific biomolecules such as lipids or sterols,” clarifies Robert Hazen from the Carnegie Institution for Science in the US, who is leading the group. “Rather, we seek subtle indications in the distribution of all the tiny molecular fragments resulting from the breakdown of the original molecules.”
Biomolecular Reverberations
To achieve this, the team initially gathered 406 varied samples, primarily sourced from the collections of distinguished palaeontologists. These samples encompassed ancient sediments, fossils, current plants, animals, fungi, and meteorites. The researchers then examined them through pyrolysis–gas chromatography–mass spectrometry. This process effectively decomposed both organic and inorganic substances contained within, releasing chemical fragments reminiscent of long-decomposed biomolecules.
The team subsequently trained a machine learning model using about 75% of the samples to discern patterns among the numerous chemical fragments, thereby determining if they were of biological or non-biological origin and whether they resulted from photosynthesis or not. The remaining 25% of samples were used for validation, demonstrating an accuracy between 90% and 100%.
Among the findings, the technique revealed that chemical fragments emitted from 3.3-billion-year-old sedimentary rock in South Africa were of biological origin, though photosynthesis-related molecules were not found. In contrast, photosynthetic molecules were detected in another South African sample approximately 2.5 billion years old, expanding the chemical record of photosynthesis by over 800 million years.
“We were amazed [by that finding],” states Hazen. “You or I could never recognize the patterns in those fragments, but AI can. The distribution of hundreds to thousands of fragments narrates the history of ancient life. My aspiration is for this method to become a standard technique in palaeobiology and astrobiology, because the identical approach can be utilized to search for life on Mars.”
“The research appears extensive and thoughtfully conceived. The machine learning approach itself isn’t new, but its application to this geochemical system is quite innovative,” remarks Tanai Cardona, who studies the origins of photosynthesis at Queen Mary University of London, UK. “The findings do not introduce a new viewpoint on the evolution of photosynthesis, but they indicate that the method can enhance and aligns with other techniques.”
Cardona believes it would be intriguing to examine a broader array of older samples from the Archean eon, which commenced 4 billion years ago. “In fact, they should evaluate all available samples to try and establish when oxygenic photosynthesis signatures first convincingly appear,” he mentions. “This would be challenging since various metabolic processes occur concurrently in many environments.”
Hazen asserts this is merely the onset. “We require thousands of diverse and well-documented samples. Many scientists have already reached out to us, offering valuable new samples from Australia, South Africa, Greenland, and Canada,” he adds. “The more varied samples we have, the more substantial the results, and the more characteristics we can extract – for instance, different types of photosynthesis or prokaryotes versus eukaryotes. The possibilities ahead are vast.”