A machine learning framework can spot molecules made by biological processes and could be used to analyze samples returned by current and future planetary missions. This research is published in PNAS Nexus.
The researchers looked at organic compounds in meteorites from space and samples from Earth like soils and rocks. They wanted to find a way to tell abiotic organics made without life and biotic organics from living things apart, especially for future space missions that might bring back samples with signs of life.
They used a method called two-dimensional gas chromatography coupled with high-resolution time-of-flight mass spectrometry. Gas chromatography separates molecules by how they move, and mass spectrometry measures their mass to identify them. This created large amounts of data on the molecules' properties like mass and movement times.
To handle this data, the researchers developed LifeTracer, a computer program that processes the information and uses machine learning, which is teaching computers to find patterns, to classify the samples.
How LifeTracer identifies origins
LifeTracer finds peaks in the data from molecules and groups them into features. It then trains a logistic regression model to separate abiotic from biotic samples. The model got over 87 percent accuracy.
The researxchers found differences, like meteorites having more polycyclic aromatic hydrocarbons, which are ring-shaped carbon molecules, and Earth samples having other structures. This helps because life makes molecules in specific ways, unlike random non-living processes.
This approach looks at all organics together, not just specific signs of life, making it less biased. It could help check samples from Mars or other places for life without missing unusual forms. The work used samples like Murchison meteorite and Iceland soil, and careful steps to clean data and avoid errors. Overall, it offers a new tool for astrobiology, the study of life in space, to better understand origins of organics in the solar system.