Dinosaurs leave behind unmistakable physical remains as fossils. It’s hard to miss, with its adult-sized femur and skull filled with serrated teeth. But for the first 90 percent of Earth’s history, life was tiny, soft, and squishy. Ancient microbes did not leave behind skeletons when they died, but some controversially argue that colonies left behind ancient microstructures. The microorganisms left behind chemical traces such as trace amounts of lipids and amino acids that became trapped in the mud and eventually turned into stone.
The problem is that the chemistry is diluted. Over billions of years, heat and pressure scramble these molecules until they become indistinguishable from the abiotic carbon, the dead, inanimate material found in meteorites. For decades, this has been the fog of war for biologists. We know that life existed 3 billion years ago, but the molecular proof is often debated or dismissed as pollution.
A new study recently published in the Proceedings of the National Academy of Sciences breaks through that fog. Using a combination of mass spectrometry and artificial intelligence, a team at the Carnegie Institution for Science has pushed back the known molecular record of photosynthesis by almost 800 million years.
The research team, led by astrobiologist Michael Wong and geologist Robert Hazen, found no new fossils. They discovered a new way of seeing things. They collected 406 samples, including modern plants, ancient coal, 3.5 billion-year-old chert, and even carbonaceous meteorites. They used a technique called pyrolysis gas chromatography-mass spectrometry (Py-GC-MS). Briefly, the sample was heated until it evaporated, breaking down the organic matter into its constituent fragments.
Until now, scientists were looking for specific biomarkers. These are single, intact molecules that scream life, like cholesterol and chlorophyll. But these fragile molecules rarely survive billions of years of geological cooking. Rather than looking for a needle in a haystack, Carnegie’s team used machine learning to examine the shape of the haystack itself.
They trained a type of machine learning model called a random forest based on these molecular fingerprints. The AI didn’t look for a single molecule. We analyzed the distribution of thousands of molecular fragments. It has learned to spot the subtle, complex patterns that distinguish the chaos of biology from the order of inanimate chemistry.
Also read: Scientifically speaking, heart disease drugs disguised as weight loss injections
The results are very powerful. When tested on modern samples comparing leaves and meteorites, the AI was 100% accurate. We maintained approximately 93% accuracy when distinguishing fossilized biological samples from abiotic meteorites and synthetic organic mixtures. But the real breakthrough came when they applied their model to Earth’s oldest and most controversial rocks.
AI has identified signs of oxygenic photosynthesis in South Africa’s Gamohan Formation. These rocks are 2.52 billion years old. Prior to this, the oldest biomolecular evidence for photosynthesis was about 1.7 billion-year-old molecules preserved in rocks. This research will help match the chemical and geological records, helping to fill in a major gap in our understanding of when the Earth began breathing.
Even more surprising, the model found evidence of life in the Josefsdal chert, a 3.33 billion-year-old rock formation. The AI examined the degraded and fragmented carbon in these stones, identifying it as the remains of living organisms and distinguishing it from the meteorite carbon that often contaminates rocks from that era.
This is important for two reasons. First, let’s rewrite the opening chapter of Life on Earth. Second, it changes the way we hunt extraterrestrials. In any science fiction or movie show, if we ever discovered extraterrestrial life, it would be microbes, not Little Green Men.
Currently, the golden rule for finding life on Mars is sample return. This includes spending billions of dollars to fly the stones to Earth for analysis in the best laboratories. This new study suggests we may not necessarily have to wait for that logistical feat.
NASA’s Curiosity rover is already equipped with a Py-GC-MS instrument, the Sample Analysis at Mars or SAM suite. Cook the rocks and analyze the gas. The problem has always been in the interpretation of that data. If Curiosity discovers carbon, is it of biological origin or just an inanimate meteorite?
Carnegie’s research shows that you don’t necessarily need a pure sample to answer that question. The AI model showed that it could distinguish between biotic carbon and abiotic carbon in carbon-rich meteorites with high accuracy. This computer brain could be uploaded to the rover. Instead of sending samples home, software could be sent to Mars, but that would require careful calibration and rigorous testing of the rover’s instruments.
Also read: Speaking scientifically: How AI is infiltrating scientific journals
Science often advances not because it discovers something new, but because it finds new ways to examine old things. We’ve been staring at these South African rocks for decades. The computer is discovering ghosts hiding within it. But it may just be the beginning. If computers can distinguish between 3 billion-year-old microbes and space rocks on Earth, it may be best to do the same on Mars.
Anirban Mahapatra is a scientist and author who recently wrote the popular science book When Medicines Don’t Work: The Hidden Pandemic That Could End Healthcare. The views expressed are personal.
