Can scientists detect life without knowing exactly what it looks like? A machine-learning approach offers a fresh perspective.
Newswise — When NASA’s OSIRIS-REx mission opened its sample-return canister in late 2023, researchers were astonished by what they found. Dust and rocks from the asteroid Bennu carried many of life’s essential ingredients: all five nucleobases used in DNA and RNA, 14 of the 20 amino acids found in proteins, and a broad array of other organic molecules. These components are primarily built from carbon and hydrogen and often form the backbone of biological chemistry.
For decades, scientists have speculated that ancient asteroids might have delivered the raw materials for life to Earth, and Bennu’s cargo seemed to support that idea. Even more surprising was the amino acids’ near-equal mix of left-handed and right-handed (chiral) forms. On Earth, virtually all biological systems rely on left-handed amino acids. If Bennu had shown a strong excess of one handedness, it could have suggested that life’s molecular asymmetry was seeded in space. Instead, the roughly balanced mixture points to a different possibility: life’s preference for left-handed amino acids likely developed later on Earth, through terrestrial processes rather than being pre-programmed in the material delivered by asteroids.
If space rocks can carry familiar ingredients without leaving a clear “life signature,” then identifying true biosignatures becomes notably more complex. This question grows even more urgent as new missions target Mars, its moons, and ocean worlds in our solar system. How can researchers detect life when chemistry alone begins to resemble lifelike organization? If nonliving materials can form rich, organized organic mixtures, the traditional signs of biology may no longer suffice.
As a computational scientist focused on biological signatures, I confront this challenge directly. In my astrobiology research, I ask how to determine whether a set of molecules arose from complex geochemistry or from extraterrestrial biology when exploring other worlds.
In a new study published in PNAS Nexus, my colleagues and I introduced a framework called LifeTracer to address this issue. Rather than chasing a single molecule or structure as proof of life, we assess how likely mixtures of compounds found in rocks and meteorites are to contain traces of life by analyzing the full chemical patterns they hold.
Identifying Potential Biosignatures
The central idea behind LifeTracer is that life creates molecules with purpose, while nonliving chemistry does not. Cells store energy, build membranes, and transmit information, whereas abiotic chemistry—driven only by nonliving processes—lacks these metabolic and evolutionary influences.
Traditional biosignature searches often aim for specific compounds (such as certain amino acids or lipid structures) or for chiral bias. While these cues can be compelling, they’re grounded in Earth-based biology and may miss alien life that uses different chemistry. They also risk mislabeling complex nonliving chemistry as evidence of life.
The Bennu results highlight this risk: a sample containing familiar life-related molecules does not, by itself, indicate biology. To reduce this bias, we built a curated dataset that sits at the boundary between life and nonlife. It combines eight carbon-rich meteorite samples, which preserve abiotic solar-system chemistry, with ten Earth-derived soils and sediments containing degraded remnants of biological molecules. Each sample cataloged tens of thousands of organic molecules, many in low abundance and many with partially unknown structures.
At NASA’s Goddard Space Flight Center, scientists processed each sample by crushing, solvent extraction, and heating—an approach reminiscent of brewing tea. The resulting mixture of organics was then filtered through two columns to separate the complex molecular blend. The organics were subsequently fragmented using electron bombardment, producing thousands of small pieces used for analysis.
Traditionally, chemists would try to reconstruct each molecule from these fragments, but with tens of thousands of compounds per sample, that task is impractical.
LifeTracer
LifeTracer represents a novel data-analytic approach: instead of rebuilding molecular structures, it analyzes the fragmentation patterns themselves. Each piece is described by its mass and two other chemical properties, and these pieces are organized into a large matrix representing the sample’s molecular landscape. A machine-learning model is then trained to distinguish meteorite samples from Earth-based terrestrial materials based on the distribution of these molecular fingerprints.
Supervised learning, a common machine-learning paradigm, uses labeled input-output pairs to learn rules. Remarkably, LifeTracer achieved strong separation between abiotic and biotic origins even with only 18 labeled examples. The key was focusing on the overall distribution of chemical fingerprints rather than chasing a single marker.
Meteorites tended to contain more volatile compounds that readily evaporate or break apart, reflecting the cold, space-based chemistry they originate from. Some molecules, such as polycyclic aromatic hydrocarbons, appeared in both groups but exhibited distinct structural differences that the model could interpret. A sulfur-containing compound called 1,2,4-trithiolane emerged as a robust marker for abiotic samples, while compounds arising from biological processes dominated the terrestrial group.
These findings suggest that the contrast between life and nonlife hinges on the arrangement and pattern of many molecules, not on a lone clue. By focusing on holistic patterns rather than preset expectations about Earth-like chemistry, LifeTracer opens new possibilities for evaluating samples returned from future missions to Mars, its moons Phobos and Deimos, and the icy worlds like Europa and Enceladus.
Looking ahead, future samples will likely contain mixed organic signatures from multiple sources. Rather than relying on a handful of familiar molecules, scientists can assess whether the overall chemical landscape resembles biology or random geochemistry.
LifeTracer is not a universal life detector. Rather, it provides a principled framework for interpreting complex organic mixtures. Bennu’s findings remind us that life-friendly chemistry may be widespread in the solar system, but chemistry alone does not prove biology.
To truly tell the difference, researchers will need every tool at their disposal—more capable spacecraft, smarter instruments, and, crucially, smarter ways to read the stories hidden in the molecules these missions return.
