As space agencies prepare to bring back parts of the Martian system, the hope of finding life somewhere in the universe feels closer than ever. NASA’s Mars Sample Return Mission and Japan’s Mars Satellite Exploration Mission aim to collect rocks and dust that may hold traces of past biology.
These materials could help answer a question that has shaped scientists for generations: whether life has taken root beyond Earth. But the challenge won’t end even if those samples land on Earth. We need a reliable way to know whether the organic molecules inside came from a living system or were formed by simple chemistry without any life at all.
A new way to see life through complex chemistry
Scientists have typically relied on a small set of reliable markers to determine whether a molecule originates from biology. The presence of particular amino acids, a bias toward single-molecule “handedness,” or complex structures often suggests life. However, this approach can fail if molecules are produced by both natural processes and organisms. There is also a risk of overlooking unknown chemistry that could signal life in a form unlike anything else on Earth.
New research suggests a different strategy. Instead of searching for a few standout compounds, the researchers created a computational framework called LifeTracer that examines the entire inventory of chemicals. By using machine learning to classify patterns among thousands of organic fragments, the system can separate biological features from non-biological chemistry with more than 87% accuracy. This study suggests that life leaves behind broad patterns in molecular distribution, not just a handful of special signals.
Build a training set from Earth and space
To develop LifeTracer, the team needed samples of both abiotic and biotic organic matter. They analyzed eight carbon-rich meteorites, including well-known samples such as Murchison and Orgueil. These space rocks contain soluble organic matter that formed in the cold space environment long before Earth existed. These are among the oldest solid materials ever studied.
The researchers then collected 10 Earth samples from locations including Antarctica, the Atacama Desert, Utah, Iceland, and Spain’s Rio Tinto region. These materials contain degraded remains of life from ancient, harsh, or biologically rare environments. Together, the two groups demonstrated a clear contrast between a chemistry shaped by life and a chemistry shaped solely by physics.
Past studies have often focused on specific molecules that differ between biological and abiotic sources. While meteorites tend to contain racemic amino acids and lack complex isoprenoids, Earth samples contain proteins with strict L-amino acid structures and long biochemical chains. However, no one had rigorously compared the complete distribution of soluble organic matter across both sets. Without that broader perspective, it was difficult to define what biochemistry and abiotic chemistry are when considered as complete systems.
Turn vast chemical situations into data
To capture these complete chemical profiles, the researchers used a combination of two-dimensional gas chromatography and high-resolution time-of-flight mass spectrometry. For each sample, the instrument measured the mass-to-charge ratio of the organic fragments, their two retention times as they moved through the instrument, and their intensity, which reflects their abundance. The resulting four-dimensional data created thousands of measurement points for each sample.
On average, meteorites contained approximately 1,184 detected peaks, and terrestrial material contained approximately 907 peaks. Each peak represents a unique fragment ion, providing a glimpse into the incredible chemical diversity within these materials.
Statistical tests showed that meteorites and terrestrial samples have distinct differences in the mass range of their fragments and the amount of time those fragments remain in the system before being detected. Organic matter from the meteorite moved faster through the device, suggesting it was lighter and more volatile molecules. These results were consistent with expectations for chemistry formed without biology.
Teach machine learning to detect life’s fingerprints
To turn this pile of chemistry into something computers could learn, the researchers grouped peaks with common mass values and similar retention times. These clusters represent fragments with similar origins and behaviors. This process generated over 9,000 features that were used to train the logistic regression model.
The model learned to classify samples as either meteorites or terrestrial with high accuracy. When tested on never-before-seen samples, it performed well with an area under the curve score of over 0.93. The researchers then refined the features by grouping them into large sets of 140 that were thought to correspond to specific types of parent compounds. Many belonged to families of molecules known to form under specific chemical conditions.
Some of the most important pro-biosignals came from compounds similar to polysubstituted alkylbenzenes and decalin derivatives. These have appeared in several Earth samples and one meteorite. On the abiotic side, naphthalene consistently emerged as the most predictive feature. Several alkylated polycyclic aromatic hydrocarbons also showed strong links with abiotic chemistry in meteorites. These molecule families repeatedly appeared among those that contributed most to LifeTracer decisions.
A wide range of ways to find life
The study’s authors note that real Mars samples may contain a mixture of organic matter from several sources, including abiotic reactions, meteorite debris, and possibly living organisms that have degraded over long periods of time. Traditional biosignatures may be missing or modified beyond recognition. LifeTracer does not require a single defining molecule. Instead, we look at the entire chemical distribution and treat the data like a fingerprint that reflects the processes that shaped it.
As the Mars mission prepares to return its first samples, this approach could help scientists determine whether a sample’s chemistry is closer to that of a meteorite or the remains of life. It also highlights compounds that are worth further study in the laboratory.
Researchers claim that combining machine learning with detailed chemical analysis may be one of the most effective ways forward. If extraterrestrial life has ever left its molecular signature in the rocks and soil of Mars, this kind of system-wide analysis may finally help us recognize it.
The study results are available online in the journal PNAS Nexus.
