Can scientists detect life without knowing what it looks like? Research using machine learning offers a new way

When NASA scientists opened the sample return canister of the OSIRIS-REx asteroid sampling mission in late 2023, they discovered something astonishing.

Dust and rocks collected from the asteroid Bennu contained many of the building blocks of life, including the five nucleic bases used in DNA and RNA, 14 of the 20 amino acids found in proteins, and a rich collection of other organic molecules. These are built primarily from carbon and hydrogen and often form the backbone of the chemistry of life.

For decades, scientists have predicted that the first asteroids could have brought the ingredients of life to Earth, and these findings seemed like promising evidence.

Even more surprising, these Bennu amino acids were distributed almost equally between the “left-handed” and “right-handed” forms. Amino acids come in two mirror image configurations, just like our left and right hands, called chiral forms.

On Earth, almost all biology requires left-handed versions. If scientists had discovered a strong excess of left-handedness in Bennu, it would have suggested that the molecular asymmetry of life may have been inherited directly from space. Instead, the nearly equal mixing indicates a different story: Life’s preference for left-handers probably arose later, through processes on Earth, rather than being pre-imprinted in the material delivered by asteroids.

If space rocks may contain familiar ingredients but not the chemical “signature” that life leaves behind, then identifying the true signs of biology becomes extremely complicated.

These findings raise a deeper question, one that becomes more pressing as new missions target Mars, the Martian moons, and the ocean worlds of our solar system: How do researchers detect life when chemistry alone starts to look “realistic”? If non-living materials can produce rich, organized mixtures of organic molecules, then the traditional signs we use to recognize biology may no longer be enough.

As a computer scientist studying biological signatures, I face this challenge firsthand. In my astrobiology work, I wonder how to determine whether a set of molecules was formed by complex geochemistry or by extraterrestrial biology, while exploring other planets.

In a new study published in the journal PNAS Nexus, my colleagues and I developed a framework called LifeTracer to help answer this question. Instead of looking for a single molecule or structure that proves the presence of biology, we attempted to classify the likelihood that mixtures of compounds preserved in rocks and meteorites contain traces of life by examining the complete chemical patterns they contain.

Identify potential biosignatures

The key idea behind our framework is that life produces molecules with a specific purpose, unlike non-living chemistry. Cells must store energy, build membranes and transmit information. Abiotic chemistry produced by nonliving chemical processes, even when abundant, follows different rules because it is not shaped by metabolism or evolution.

Traditional biosignature approaches focus on finding specific compounds, such as certain amino acids or lipid structures, or chiral preferences, such as left-handedness.

These signals can be powerful, but they rely entirely on the molecular patterns used by life on Earth. If we assume that extraterrestrial life uses the same chemistry, we risk missing biology similar – but not identical – to ours, or misidentifying non-living chemistry as a sign of life.

The Bennu results highlight this problem. The asteroid sample contained molecules familiar to life, but nothing inside appears to have been alive.

To reduce the risk of assuming that these molecules indicate life, we assembled a unique dataset of organic materials right on the border between life and non-life. We used samples from eight carbon-rich meteorites that preserve the abiotic chemistry of the early solar system, as well as 10 samples of Earth’s soil and sedimentary materials, containing the degraded remains of biological molecules from past or present life. Each sample contained tens of thousands of organic molecules, many of which were present in low abundance and whose structure could not be fully identified.

At NASA’s Goddard Space Flight Center, our team of scientists ground each sample, added solvent, and heated it to extract the organic material – this process is like brewing tea. Next, we took the “tea” containing the extracted organic materials and passed it through two filter columns that separated the complex mixture of organic molecules. Then the organic materials were pushed into a chamber where we bombarded them with electrons until they broke into smaller fragments.

Traditionally, chemists use these mass fragments as puzzle pieces to reconstruct each molecular structure, but having tens of thousands of compounds in each sample presented a challenge.

life tracer

LifeTracer is a unique approach to data analysis: it works by taking the fragmented pieces of the puzzle and analyzing them to find specific patterns, rather than reconstructing each structure.

He characterizes these puzzle pieces by their mass and two other chemical properties, then organizes them into a large matrix describing all of the molecules present in each sample. It then trains a machine learning model to distinguish meteorites from terrestrial materials on the Earth’s surface, based on the type of molecules present in each.

One of the most common forms of machine learning is called supervised learning. It works by taking many pairs of inputs and outputs as examples and learns a rule to move from input to output. Even with only 18 samples as examples, LifeTracer performed remarkably well. It systematically separates abiotic origins from biotic origins.

What mattered most to LifeTracer was not the presence of a specific molecule but the overall distribution of chemical fingerprints found in each sample. Meteorite samples tended to contain more volatile compounds – they evaporate or break apart more easily – reflecting the type of chemistry most common in the cold environment of space.

Certain types of molecules, called polycyclic aromatic hydrocarbons, were present in both groups, but they had distinctive structural differences that the model could analyze. A sulfur-containing compound, 1,2,4-trithiolane, emerged as a strong marker for abiotic samples, while terrestrial materials contained products formed by a biological process.

These discoveries suggest that the contrast between life and non-life is not defined by a single chemical index but by the way in which a whole set of organic molecules is organized. By focusing on models rather than hypotheses about which molecules life “should” use, approaches such as LifeTracer open new possibilities for evaluating samples returned by missions to Mars, its moons Phobos and Deimos, Jupiter’s moon Europa, and Saturn’s moon Enceladus.

Future samples will likely contain mixtures of organic materials from multiple sources, some organic and some not. Instead of relying only on a few familiar molecules, we can now assess whether the entire chemical landscape is more like biology or random geochemistry.

LifeTracer is not a universal life detector. Rather, it provides a basis for the interpretation of complex organic mixtures. Bennu’s findings remind us that life-friendly chemistry may be widespread throughout the solar system, but chemistry alone does not equal biology.

To make a difference, scientists will need all the tools we can build – not only better spacecraft and instruments, but also smarter ways to read the stories written in the molecules they bring home.

This article is republished from The Conversation, an independent, nonprofit news organization that brings you trusted facts and analysis to help you make sense of our complex world. It was written by: Amirali Aghazadeh, Georgia Institute of Technology

Learn more:

Amirali Aghazadeh receives funding from Georgia Tech.