Detecting signs of life in biotic–abiotic mixtures using pyrolysis–gas chromatography–mass spectrometry and machine learning

Biosignature detection is at the forefront of modern astrobiology and paleobiology research. Previous studies have demonstrated that machine learning models trained on data from pyrolysis–gas chromatography–mass spectrometry (py–GC–MS) can distinguish purely biotic and abiotic specimens from one another, as well as identify differences among contemporary biotic and ancient biotic samples. However, if life exists beyond Earth, its organics may be mixed with abiotic organics from, e.g. , those derived from meteoritic infall. Thus, it is critical to test whether mixtures containing both biotic and abiotic components can be accurately characterized by machine learning–based techniques. Here, we investigate whether machine learning models can accurately determine the biogenicity of mixed biotic-abiotic samples. We used py–GC–MS to analyze the organic content of 61 mixtures, which included various combinations of living organisms, fossilized organics, carbonaceous meteorites, and laboratory-synthesized organics. Different ratios of the components in each mixture are used to examine how much biological organic materials can be detected in a sample. The ratios (% component 1:% component 2) include 90:10, 66:33, 50:50, 33:66, and 10:90. The data from the chromatographic and mass peaks produced by the mixed samples were then analyzed by a random forest model trained on purely biotic and purely abiotic samples. This method was able to correctly identify 87% of the mixtures that contain contemporary/modern biotic components as living, with a limit of detection between about 1% and 10% biotic material. Mixtures containing ancient/old biotics were correctly identified 60% of the time, with a limit of detection between about 33% and 50% biotic material.