How can we detect life without knowing what it looks like? Machine learning offers a new approach

If nonliving chemistry can produce rich and highly ordered mixtures of organic molecules, traditional signs of biology may no longer be sufficient. This is the conclusion discussed by Space.com, referring to a new study published in PNAS Nexus.

When NASA scientists opened the container holding samples from the asteroid Bennu, returned by the OSIRIS REx mission in late 2023, they encountered an unexpected result. The 120 grams of dust and rock contained all five nucleobases of DNA and RNA, 14 of the 20 amino acids used in proteins, and many other organic compounds that serve as the building blocks of life. These molecules, composed mainly of carbon and hydrogen, form the foundation of biochemistry.

For decades, scientists have suggested that asteroids delivered the ingredients of life to Earth, and the Bennu findings strongly support this idea. Even more striking was the fact that the amino acids were almost evenly split between left handed and right handed chiral forms. On Earth, nearly all life uses only the left handed version. The balance observed in Bennu indicates that biological asymmetry emerged later on Earth itself rather than being imported from space.

If space rocks can carry familiar molecules without the chemical signature of life, the challenge of identifying true biosignatures becomes extremely difficult.

Life or just complex chemistry

How can biology be distinguished from geochemistry when the molecules themselves appear life like. This question lies at the center of a new study led by a computer scientist and astrobiologist working on biosignatures. Instead of searching for specific molecules or structures, the team developed a machine learning framework called LifeTracer, designed to classify chemical mixtures based on the likelihood that they contain traces of life.

The underlying idea is that life produces molecules with purpose, such as storing energy, building membranes, or transmitting information. Nonliving chemistry, even when highly complex, follows different rules and is not shaped by metabolism or evolution.

Traditional methods focus on identifying particular compounds like amino acids or lipids, or on detecting chiral asymmetry. These approaches work well for Earth life but may fail to detect unfamiliar forms of biology or may mistake nonliving chemistry for a biosignature.

The Bennu results highlight this problem clearly. All the ingredients are present, but there is no life.

How LifeTracer works

The research team assembled a unique dataset consisting of eight carbon rich meteorites, representing abiotic chemistry from the early Solar System, and ten samples of terrestrial soils and sedimentary rocks, which contain residues of biological molecular activity.

Each sample was ground up, organic material was extracted with a solvent in a process similar to brewing tea, the extracts were separated into fractions, and then analyzed using mass spectrometry. Each sample yielded tens of thousands of molecular signals, many present in low concentrations and many not individually identified.

LifeTracer does not attempt to reconstruct the structure of every molecule. Instead, it analyzes mass fragments based on their mass and two chemical properties, builds a large data matrix, and trains a model to distinguish meteorites from Earth samples influenced by biology.

Despite the small dataset of only 18 samples, the model achieved high accuracy. The key lies not in individual molecules but in the overall distribution of chemical fingerprints. Meteorites contain more volatile compounds typical of cold space environments. Polycyclic aromatic hydrocarbons appear in both groups but with different structural characteristics. The compound 1,2,4 trithiolane emerged as a strong marker of nonliving samples, while Earth samples were enriched in products of biological processes.

What this means for the search for life

LifeTracer is not a universal life detector but a tool for interpreting complex organic mixtures. It allows scientists to assess whether the chemistry of a sample looks more like biology than like random geochemistry.

Future missions to Mars, Phobos, Deimos, Europa, and Enceladus will return samples containing organic material from many different sources. LifeTracer could help distinguish biological signals from abiotic ones without relying on assumptions based on Earth life.

The Bennu results serve as a reminder that life like chemistry may be widespread throughout the Solar System, but chemistry alone is not the same as life.

In brief

Bennu showed that asteroids can carry all the components of RNA and proteins without any signs of life. Traditional biosignatures may therefore be insufficient. LifeTracer introduces a machine learning approach that analyzes the overall chemical profile of a sample rather than individual molecules. By comparing distributions of compounds, it can distinguish abiotic meteorite chemistry from Earth based biological chemistry. This opens the door to a more objective search for life on Mars and the moons of Jupiter and Saturn, without assuming that extraterrestrial life must resemble life on Earth.