AI Breakthrough Reveals Chemical 'Ghosts' of Ancient Life in 3.3-Billion-Year-Old Rocks

AI spots 'ghost' signatures of ancient life on Earth

In searching for the earliest life on Earth and other worlds, researchers normally look for intact fossils or biomolecules made only by living organisms. But such signals are few and far between. Now, researchers have devised an artificial intelligence (AI) that can identify signs of ancient life in rocks of unknown provenance, based only on the pattern of chemicals left behind as biomolecules degrade over eons. "We have a way to read molecular 'ghosts' left behind by early life," says Robert Hazen, a geologist at the Carnegie Institution for Science who led the study, published today in the Proceedings of the National Academy of Sciences. Using this automated pattern recognition technique, the researchers say they can see life's signature in 3.3-billion-year-old rocks, which is hundreds of millions of years shy of the indication of life in Earth's oldest fossils. But the new work also claims to push back the biomolecular signature of the earliest photosynthetic life some 800 million years to 2.5 billion years ago. Researchers are now working to adapt this approach to search for signs of life on Mars and moons of Jupiter and Saturn. "This could pan out to be very, very important," says Karen Lloyd, a microbial biogeochemist at the University of Southern California who was not involved in the study. "It's a great way to look for biosignatures." Still-debated microfossil evidence for the earliest life on Earth dates back more 3.7 billion years ago, in the form of rocks containing filaments made by microbes living around hydrothermal vents in what's now Canada. In addition, mats of bacteria in what's now Western Australia laid down more conclusive fossil evidence in mounded structures called stromatolites some 3.5 billion years ago. But such fossils from Earth's youth are extremely rare. Researchers have tried to fill out that record by searching for ancient sediments that contain not fossils, but chemical and molecular signs of life. Only living organisms are thought to be capable of making certain lipids and ring-shaped compounds called porphyrins, for example. But Earth's tectonic machinery tends to obliterate such signs, by burying, crushing, heating, and cooling the sediments. Indirect measures also offer clues. For instance, rocks more than 3.7 billion years old are enriched with carbon-12, a light isotope that's preferred by living organisms, compared with the heavier isotope carbon-13. Still, finding conclusive molecular biosignatures "has not proven an easy problem at all," says Woodward Fischer, a geobiologist at the California Institute of Technology. So, Hazen and his colleagues decided to drop the search for intact, biomolecular smoking guns. Instead, they wondered whether they could spot telltale patterns in the molecular detritus these compounds leave behind as they break down. To do so, the team amassed more than 400 samples. Some were rock and sediment samples known to contain either living or fossil organisms. Others were abiotic samples from meteorites. The team analyzed them with a tool called a pyrolysis gas chromatograph mass spectrometer (GC-MS). The device heated the samples to more than 600°C, breaking them apart into volatile fragments. The fragments were then separated by their physical and chemical properties, identified, and tallied by their concentration. Carnegie astrobiologist Michael Wong, the study's first author, likens the instrument to "a really fancy oven that not only bakes your cake, but tastes it for you, too." In the end, each sample was transformed into a landscape of data, with up to hundreds of thousands of individual peaks that each represented a different possible molecular fragment. They then used a conventional machine learning technique, known as a random forest model, to look for patterns in both what was present and what was missing. "What the machine learning model does is essentially try to use every single one of those data landscapes as a fingerprint to find what is similar to each other and what is different," says Carnegie geoinformatics expert Anirudh Prabhu. After using 75% of the samples to train their AI, the researchers then let it loose on the rest. For test samples, the AI correctly distinguished between biological and abiotic samples with more than 90% accuracy. It also saw chemical patterns unique to biology in rocks as old as 3.3 billion years, nearly twice as old as previous biomolecular signatures preserved in ancient rocks. In addition, the AI teased out the molecular pattern associated with oxygen-producing photosynthesis in rocks up to 2.5 billion years old. Although there is plentiful geochemical evidence for photosynthetic life around that time from the sudden explosion of oxygen it produced, preserved evidence of these organisms' molecular machinery is scant. The new results push back the molecular signature of photosynthetic life by more than 800 million years, the authors say. Not all signatures were easy to see. For samples believed to be biotic, in rocks 500 million to 2.5 billion years old, the AI identified signatures of life about two-thirds of the time. But in rocks older than 2.5 billion years, that number dropped to 47%. For each sample, the model didn't just report whether life's signature was present, it also provided a probability score. If a sample scored above 60% for "biotic," it was considered a strong hit. "The confidence is not as good as you'd want it to be," Lloyd says. Still, she notes, that could change as researchers bolster the AI's training data with more samples. The researchers are eager to test the system on extraterrestrial samples. According to Prabhu, the model "opens the door to exploring ancient and alien environments with a fresh lens, guided by patterns we might not even know to look for ourselves." Wong adds that this biosignature pattern recognition should also work with other analytical tools, which could help future robotic missions to Mars, Jupiter's moon Europa, and Saturn's moon Enceladus broaden their search for signatures of extraterrestrial life. Wong's group is starting a new, $5 million, NASA-funded project to do just that. The goal, he says, is to "answer one of the greatest scientific questions we still have left to answer, which is: Are we alone in the universe?"

Secret chemical traces reveal life on Earth 3. 3 billion years ago

Researchers from the Carnegie Institution for Science led an international effort that combined state-of-the-art chemical techniques with artificial intelligence. Their goal was to uncover extremely subtle chemical "whispers" of past biology hidden inside heavily altered ancient rocks. By applying machine learning, the team trained computer models to recognize faint molecular fingerprints left by living organisms long after the original biomolecules were destroyed. Seaweed Fossils Offer a Window Into Early Complex Life Michigan State University's Katie Maloney, an assistant professor in the Department of Earth and Environmental Sciences, contributed to the project. Her work focuses on how early complex life evolved and shaped ancient ecosystems. Maloney provided exceptionally well-preserved seaweed fossils that are roughly one billion years old, collected from Yukon Territory, Canada. These fossils are among the earliest known seaweeds in the geological record, dating to a time when most organisms were visible only under a microscope. The study, published in the Proceedings of the National Academy of Sciences, offers new understanding of Earth's earliest biosphere. It also carries major implications for exploring life beyond Earth. The same methods could be applied to samples from Mars or other planetary bodies to determine whether they once supported life. "Ancient rocks are full of interesting puzzles that tell us the story of life on Earth, but a few of the pieces are always missing," Maloney said. "Pairing chemical analysis and machine learning has revealed biological clues about ancient life that were previously invisible." Why Early Biosignatures Are So Hard to Find Life on early Earth left behind only sparse molecular evidence. Fragile materials such as primitive cells and microbial mats were buried, squeezed, heated, and fractured as the planet's crust shifted over billions of years. These intense processes destroyed most original biosignatures that could have provided insight into life's earliest stages. Yet the new findings show that even after original molecules vanish, the arrangement of surviving fragments can still reveal important information about ancient ecosystems. This research demonstrates that ancient life left behind more signals than scientists once suspected -- faint chemical "whispers" preserved within the rock record. To identify these clues, the team used high-resolution chemical techniques to break down both organic and inorganic material into molecular fragments. They then trained an artificial intelligence system to recognize the chemical "fingerprints" associated with biological origins. The researchers analyzed more than 400 samples, ranging from modern plants and animals to billion-year-old fossils and meteorites. The AI system distinguished biological from nonbiological materials with over 90 percent accuracy and detected signs of photosynthesis in rocks at least 2.5 billion years old. Doubling the Time Span for Detecting Ancient Life Before this work, dependable molecular evidence for life had only been identified in rocks younger than 1.7 billion years. This new approach effectively doubles the period during which scientists can study chemical biosignatures. "Ancient life leaves more than fossils; it leaves chemical echoes," said Dr. Robert Hazen, senior staff scientist at Carnegie and a co-lead author. "Using machine learning, we can now reliably interpret these echoes for the first time." A New Way to Explore Earth's Deep Past and Other Worlds For Maloney, who studies how early photosynthetic organisms reshaped the planet, the results are especially meaningful. "This innovative technique helps us to read the deep time fossil record in a new way," she said. "This could help guide the search for life on other planets."

AI Uncovers Evidence of Life in 3.3-Billion-Year-Old Rocks

Earth is roughly 4.5 billion years old. Thanks to a wealth of indirect evidence from isotopes, stromatolites, and microfossils, scientists believe life emerged around 3.7 billion years ago. But direct evidenceâ€"the biochemical recordâ€"only dates back about 1.6 billion years. By combining machine learning with cutting-edge chemical analysis, an international team of researchers has detected biosignatures that significantly extend Earth's biochemical record, corroborating what indirect evidence has long suggested. The findings, published Monday in the journal Proceedings of the National Academy of Sciences, include the detection of chemical “fingerprints†left behind by microbes in 3.3-billion-year-old rocks. On top of this, they found chemical signatures of photosynthetic life in rocks as old as 2.5 billion years, extending the chemical record of photosynthesis preserved in carbon molecules by over 800 million years. “Scientists have developed many different ways to infer life in ancient samplesâ€"looking at the textures of rocks, their minerals, the isotopesâ€"but using complex molecules to come up with an unambiguous record of life only extended previously to about 1.6 billion years ago,†co-lead author Michael L. Wong, a research scientist at Carnegie Science’s Earth & Planets Laboratory, told Gizmodo. “We're taking that all the way to 3.3 [billion], so doubling that age.†Wong led the project alongside Anirudh Prabhu, another research scientist at Carnegie Science’s Earth & Planets Laboratory. While Wong specializes in astrobiology and planetary science, Prabhu is an AI and machine learning expert. To understand how this new model accurately distinguishes biosignatures from abiotic materials, you can think of it like facial-recognition software, Prabhu told Gizmodo. The model is trained on GC-MS (gas chromatography mass spectrometry) data. This 3D spectral data looks kind of like a landscape, with peaks, valleys, hills, and other features, Prabhu explained. The model identifies patterns among these features that correspond to biological materials, similarly to how facial recognition software is trained to identify the shapes that make up a person’s eyes, mouth, nose, and bone structure. “We're looking at the entire data[set], and the model is able to pick out specific features that are very key to a sample being photosynthetic or notâ€"or biogenic or notâ€"in a manner that humans just can't do because of how vast the data is,†Prabhu explained. The model is currently able to do this with 90% accuracy, and the researchers hope it will improve as it trains on more data from an increasingly diverse set of samples. This new technique could be a game changer for paleobiologists, allowing them to detect ancient biomarkers even in badly degraded or deformed samples. It’s already opening up a whole new world of opportunity for ancient chemical analysis, and Earth is only just the beginning. The search for ancient life extends far beyond our home planet. Astrobiologists like Wong look for evidence of life elsewhere in the solar system, such as on Mars or Saturn’s icy moons. The fact that the AI was able to accurately detect signs of ancient life on Earth “boosted my confidence that we’re on the right track for developing the kinds of instrumentation and machine learning algorithms that we need to try to find evidence of life in, say, ancient Mars rocks,†Wong said. “I’m full of optimism for the applications elsewhere, beyond Earth.†Wong, Prabhu, and their colleagues chose to train the AI on GC-MS data largely because it is a flight-ready instrument. “It has spaceflight heritage, there’s one of these pyrolysis GC-MS instruments sitting in the belly of the Curiosity rover on Mars right now,†Wong said. The model’s design also prioritizes computational lightweightness and interpretability, which is critical for conducting analyses in real-time as rovers collect geological samples, Prabhu explained. “So you have a rover on Mars or some other planet, it picks up a sample, zaps it, and produces the spectra. You can quickly get a preliminary predictionâ€"a highly accurate, but preliminary predictionâ€"that scientists can use to understand that area and make decisions,†he said. Both Wong and Prabhu hope to see this technology applied across the solar system, and they’ll be seeking NASA partnerships to expand its capabilities and ultimately send it to space. For now, the model will continue to deepen our understanding of the emergence of life on Earth, helping us unravel the mysteries of our origin.

Life may have emerged a billion years earlier than we thought

Earth holds memories far older than humans. These memories rest inside ancient stones shaped by heat, pressure, and time. Many early signs of life vanished as these rocks changed deep within the crust. Yet scientists continue to search for faint signals that survived. A new study reveals chemical traces of biology in rocks older than 3.3 billion years. The research suggests that oxygen-producing photosynthesis began far earlier than expected. The study paired modern chemistry with artificial intelligence. The goal was simple: to read chemical messages left behind by early organisms. The messages exist as small molecular fragments. These fragments persist even when original cells or biomolecules have long disappeared. The team trained machine learning systems to study the chemical profiles of many organic materials. The system learned how life shapes molecular patterns in a way that non biological processes do not. Once trained, the system examined ancient rocks and detected signals that point to life, even in samples older than three billion years. The study shows that life left a more durable chemical trail than once thought. These trails appear as faint patterns inside highly altered rocks. Many earlier models could not read these traces. The new method opens a wider window into early Earth, building on a growing body of work that uses chemical data and artificial intelligence. Previous research showed that organic matter formed by life carries different molecular patterns than material formed through non biological chemistry. The new study extends that idea with a larger range of samples and a more powerful model. Michigan State University scientist Katie Maloney joined the project. She studies early complex life and ancient ecosystems. Maloney contributed rare seaweed fossils from Canada that date back one billion years. These fossils represent some of the earliest known seaweeds at a time when most life stayed microscopic. "Ancient rocks are full of interesting puzzles that tell us the story of life on Earth, but a few of the pieces are always missing," Maloney said. "Pairing chemical analysis and machine learning has revealed biological clues about ancient life that were previously invisible." Her samples helped confirm that the method works on very old fossils as well as younger ones. These fossils hold organic fragments that still reflect the nature of the organisms that once lived. The new study was focused on detailed analyses of more than 400 samples. These samples included modern plants, modern animals, fossil microbes, fossil plants, meteorites, ancient sedimentary rocks, and synthetic materials. The researchers examined each sample using pyrolysis gas chromatography with mass spectrometry. This method breaks organic matter into many small fragments for study. Artificial intelligence then examined these fragments. The models separated biological and non biological material with strong accuracy. The models also identified traits linked to metabolism. These traits included signals from photosynthetic organisms as well as non photosynthetic organisms. The results confirm that even broken and degraded molecular remains still hold a clear biological pattern. Heat, pressure, and chemical changes often destroy full biomolecules. Yet the pattern formed by many fragments still reveals the presence of life. The team found that several samples older than three billion years carry strong signals of biological origin. Some samples resemble microbial material. Others show clear links to early forms of photosynthetic life. This discovery pushes back the evidence for oxygen-producing photosynthesis by nearly a billion years. The method also distinguishes biological organic matter from meteorite derived organic matter. This is important because some ancient rocks contain both types. The ability to separate the two offers strong potential for future space missions. Researchers need to know if a rock from Mars or another world once held living organisms or only non biological carbon from space. "Ancient life leaves more than fossils; it leaves chemical echoes," said Dr. Robert Hazen. "Using machine learning, we can now reliably interpret these echoes for the first time." The combined findings show a pattern through Earth history. Younger rocks contain clear biological signatures. Older rocks contain weaker but still recognizable biological signals. This decline likely reflects increasing molecular damage with age. It may also reflect the presence of some non biological organic matter in the oldest samples. Yet the important point remains. Many Paleoarchean samples still carry patterns that point to life. These results align with earlier work that used morphology and isotopes to study early life. The new method adds chemical strength to those earlier lines of evidence. Together, they show that early Earth hosted a rich microbial world long before complex organisms appeared. "This innovative technique helps us to read the deep time fossil record in a new way," said Maloney. "This could help guide the search for life on other planets." The method may soon combine more types of data. Scientists plan to add isotope ratios, Raman spectra, infrared spectra, and morphological data. Such additions may reveal even deeper details about early life. They may also help detect non-oxygen-based photosynthesis and other ancient metabolic processes. The long term goal is clear. Researchers want to understand how life began, how it changed through deep time, and how to detect it on distant worlds. The new results show that even Earth's oldest rocks still preserve chemical memories of early life - and those memories are now speaking more clearly than ever. The study is published in the journal Proceedings of the National Academy of Sciences. -- - Like what you read? Subscribe to our newsletter for engaging articles, exclusive content, and the latest updates.