AI framework studies entire chemical profiles to spot signs of life in Martian samples with high accuracy. (CREDIT: NASA/JPL-Caltech)

As space agencies prepare to bring home pieces of the Martian system, the hope of finding life elsewhere in the universe feels closer than ever. NASA’s Mars Sample Return plan and the Martian Moons eXploration mission from Japan aim to collect rocks and dust that may hold traces of past biology.

Those materials could help answer a question that has shaped generations of scientists: whether life ever took root beyond Earth. Yet the challenge does not end when those samples land on this planet. You will need a reliable way to tell if the organic molecules inside came from living systems or if they formed through simple chemistry without life at all.

A New Way to See Life in Complex Chemistry

Scientists have typically relied on a small set of trusted markers to judge whether a molecule came from biology. Certain amino acids, a bias toward one molecular “handedness,” or the presence of complex structures often point to life. But this approach can fail when molecules can be made by both natural processes and living organisms. It also risks overlooking unfamiliar chemistry that could signal life with forms unlike anything on Earth.

The LifeTracer workflow for collecting, curating, and analyzing the mass spectrometry data and developing a machine learning model for classifying samples. (CREDIT: PNAS Nexus)

A new study offers a different strategy. Instead of searching for a few standout compounds, researchers created a computational framework called LifeTracer that examines entire chemical inventories. By using machine learning to sort patterns among thousands of organic fragments, the system can separate biological signatures from abiotic chemistry with more than 87 percent accuracy. The work suggests that life leaves wide patterns in molecular distributions, not just a handful of special signals.

Building a Training Set from Earth and Space

To develop LifeTracer, the team needed examples of both abiotic and biotic organics. They analyzed eight carbon-rich meteorites, including well-known samples such as Murchison and Orgueil. These space rocks contain soluble organics formed in cold cosmic environments long before Earth existed. They are among the oldest solid materials ever studied.

The researchers then gathered ten Earth samples from places such as Antarctica, the Atacama Desert, Utah, Iceland, and the Rio Tinto region in Spain. These materials contain degraded remains of life from ancient, harsh, or biologically sparse environments. Together, the two groups offered a sharp contrast between chemistry shaped by life and chemistry shaped by physics alone.

Past studies often focused on certain molecules that differ between living and nonliving sources. Meteorites tend to contain racemic amino acids and lack complex isoprenoids, while Earth samples include proteins with strict L-amino acid structures and long biochemical chains. But no one had closely compared the full distributions of soluble organics across both sets. Without that broader view, it was hard to define what biotic and abiotic chemistry each look like when considered as complete systems.

Comparison of mass-to-charge ratio (m/z), first retention time (RT1), and second retention time (RT2) distributions between meteoritic (abiotic, top) and terrestrial (biotic, bottom) fragment ions. (CREDIT: PNAS Nexus)

Turning Vast Chemical Landscapes into Data

To capture those full chemical profiles, the researchers used two-dimensional gas chromatography paired with high-resolution time-of-flight mass spectrometry. For each sample, the instrument measured the mass-to-charge ratio of organic fragments, their two retention times as they moved through the equipment, and their intensity, which reflects abundance. The resulting four-dimensional data created thousands of measured points for every sample.

Meteorites contained an average of about 1,184 detected peaks, and the terrestrial materials contained roughly 907 peaks. Each peak represents a unique fragment ion, offering a glimpse of the incredible chemical diversity within these materials.

Statistical tests showed that meteorites and Earth samples had clear differences in the mass ranges of their fragments and in how long those fragments stayed in the system before detection. Organics from the meteorites moved through the instrument more quickly, hinting at lighter and more volatile molecules. These results matched expectations for chemistry shaped without biology.

Teaching Machine Learning to Detect Life’s Fingerprints

To turn this mountain of chemistry into something a computer could learn from, the researchers grouped peaks with shared mass values and similar retention times. These clusters represent fragments with similar origins and behavior. That process produced more than nine thousand features used to train a logistic regression model.

Visualization of the distribution of compounds in meteoritic samples and terrestrial geologic samples and the regression coefficients of the logistic regression model trained in LifeTracer. (CREDIT: PNAS Nexus)

The model learned to classify samples as meteoritic or terrestrial with strong accuracy. When tested on samples it had not seen before, it performed well, showing an area-under-curve score above 0.93. The researchers then refined the features by grouping them into 140 larger sets that likely corresponded to specific types of parent compounds. Many belonged to families of molecules known to form under specific chemical conditions.

Some of the most important biotic-leaning signals came from polysubstituted alkylbenzenes and compounds similar to decalin derivatives. These appeared in several Earth samples and in one meteorite. On the abiotic side, naphthalene consistently emerged as the most predictive feature. Several alkylated polycyclic aromatic hydrocarbons also showed strong ties to abiotic chemistry in meteorites. These families of molecules appeared again and again among the top contributors to LifeTracer’s decisions.

A Broader Way to Search for Life

The study’s authors note that real Martian samples may hold mixtures of organics from several sources, including abiotic reactions, meteorite fragments, and possibly long-degraded biology. Traditional biosignatures may be scarce or too altered to recognize. LifeTracer does not need any single defining molecule. Instead, it looks at the whole distribution of chemistry, treating the data like a fingerprint that reflects the processes that shaped it.

As Mars missions prepare to return their first samples, this approach could help scientists decide whether a sample’s chemistry looks closer to meteorites or closer to the remains of life. It also highlights compounds that deserve closer study in the lab.

The researchers argue that combining machine learning with detailed chemical analysis may be one of the most effective paths forward. If alien life ever left its molecular mark on Martian rocks or soils, this type of systemwide analysis may be what finally helps you recognize it.

Research findings are available online in the journal PNAS Nexus.

Related Stories

• If aliens were sending us signals, this is what they might look like
• Scientists detect 8 strange alien signals originating from nearby stars
• Galileo Project revolutionizes search for alien life through Artificial Intelligence

Like these kind of feel good stories? Get The Brighter Side of News’ newsletter.