When did life first emerge on Earth? Artificial intelligence could possibly help answer this question. Researchers have used machine learning to train a model to recognize subtle molecular fingerprints of living organisms. In this way, the AI identified traces of life in 3.33 billion-year-old rock samples – the oldest molecular evidence of life to date. The results also suggest that organisms were already carrying out photosynthesis 2.52 billion years ago, around 800 million years earlier than previously thought. The technology could also help find life on other planets.
Living cells produce typical chemical compounds that do not occur in non-living material. For paleontologists, such chemical signatures in ancient sediments can provide valuable clues about the early days of life. But these molecules also change over the course of millions of years. The oldest complex biological molecules that have so far been clearly identified as such are around 1.6 billion years old. However, other evidence, including microfossils and traces in rocks that probably come from microbes, suggests that life on Earth is more than twice as old. The oldest, albeit controversial, evidence of life is around four billion years old.
Fingerprints of biological molecules
A team led by Michael Wong from the Carnegie Institution for Science in Washington has now used a new method to search for the earliest biological molecules on our planet. The researchers trained an AI model to recognize subtle molecular fingerprints of living organisms. The idea behind it: Even if the original biomolecules, such as the components of an ancient cell membrane, have long since decayed, their molecular fragments can still leave behind specific signatures.
To train and subsequently test the AI, the researchers used 406 samples, which included both modern and fossil animals, plants and fungi, as well as meteorite rocks, synthetic organic materials and ancient sediments, of which it was unclear whether they were formed by biological or abiotic processes. Using pyrolysis-gas chromatography-mass spectrometry, the team analyzed the individual chemical components of the samples and provided the results to the AI model. “Ancient life leaves more than just fossils, it leaves chemical traces,” says Wong’s colleague Robert Hazen. “Using machine learning, we can now reliably interpret these traces for the first time.”
Machine search for traces
In fact, the AI recognized with up to 98 percent accuracy whether a sample had a biological origin or whether it came from, for example, a meteorite. In addition, she was able to distinguish between plants and animals in modern samples with a hit rate of 95 percent. This distinction was more difficult for her with older samples because there were only a few animal fossils in the training data set from which she could have learned patterns. The AI system was able to detect with an accuracy of 93 percent whether an organism once carried out photosynthesis or not. After the researchers had demonstrated how reliably the model worked using known samples, they applied it to sedimentary rocks where it was unclear whether there were traces of life hidden in them.
To Wong and his team’s surprise, using 3.33-billion-year-old rocks from the Josefsdal Chert in South Africa, the AI came to the conclusion that the molecules they contained most likely came from living organisms. If this assignment is correct, it would be the earliest detection of biological molecules on Earth. The AI model also found evidence of photosynthetic activity in 2.52 billion-year-old samples from the Gamohaan Formation in South Africa. “Understanding when photosynthesis emerged helps explain how Earth’s atmosphere became oxygen-rich – an important milestone that enabled the evolution of complex life, including humans,” says Wong.
From the researchers’ perspective, their method can not only provide new insights into the early days of the Earth, but can also be used in the search for life on other planets. “The exciting thing is that this approach does not rely on finding recognizable fossils or intact biomolecules,” explains Wong’s colleague Anirudh Prabhu. “AI has not only helped us analyze data faster, but also enabled us to make sense of confusing, degraded chemical data. It opens the door to exploring ancient and alien environments with a new perspective, guided by patterns we might not even look for ourselves.”
Source: Michael Wong (Carnegie Institution for Science, Washington, USA) et al., Proceedings of the National Academy of Sciences, doi: 10.17605/OSF.IO/G93CS