Researchers train an AI system to find extraterrestrial life

Machine learning model touted as ideal for finding LGMs on Mars, and beyond

Computer scientists have trained a machine learning model to predict whether materials contain biosignatures in the hopes that the system can be used to detect life in Martian rocks.

The algorithm analyzes data obtained from pyrolysis–gas chromatography mass spectrometry experiments to inspect the chemical composition of samples. The team of researchers, led by the Carnegie Institution of Science, believe that biological matter carries telltale characteristics that machine learning models can learn to identify.

"The search for extraterrestrial life remains one of the most tantalizing endeavors in modern science," Jim Cleaves, lead author of the research published in the Proceedings of the National Academy of Sciences, said in a statement. 

"The implications of this new research are many, but there are three big takeaways: First, at some deep level, biochemistry differs from abiotic organic chemistry; second, we can look at Mars and ancient Earth samples to tell if they were once alive; and third, it is likely this new method could distinguish alternative biospheres from those of Earth, with significant implications for future astrobiology missions."

To build the model the boffins first collected data drawn from analysis of 134 materials, including 75 made up of non-living matter like meteorites or polymers. The scientists also considered material from living things, such as hair, rice, microbes, oil, and fossils. Next, they split the dataset and trained their classifier algorithm on data from 95 samples, and tested its performance on the remaining 39; it achieved an average of 90 percent accuracy.

The results, however, should be taken with a pinch of salt. All the biotic molecules analyzed come from material that are of Earthly origin all, and it's not clear whether the algorithm will be effective at assessing extraterrestrial material. The researchers, however, believe that alien life will still exhibit "molecular frequency distributions that are distinct from those of background abiotic synthetic processes".

"Rather than a particular molecule or peak, it is the presence of a set of peaks across a span of the chromatograph," Cleaves told The Register. "We think this is a marker of the diversity of molecule types with respect to polarity/hydrophobicity in living organisms as opposed to abiological samples."

What those distributions and processes might look like in pyrolysis–gas chromatography mass spectrometry data remains unknown.

Although scientists have collected ancient organic samples from objects like meteorites or asteroids, it's difficult to tell whether they contain any biotic molecules since they degrade over time. Still, the researchers want to test their software and hope to study data sampled from 3.5-billion-year-old sediments from Western Australia, rocks from Northern Canada, South Africa, and China, and meteorites from Mars.

"We are now turning our attention to a variety of controversial ancient terrestrial samples whose biogenicity is hotly debated. We hope our new method may help tip the scales for some of these," Cleaves added.

The ultimate test will be samples that scientists have confirmed contain alien biosignatures, like rocks drilled by NASA's Perseverance rover, that the space agency hopes to send back to Earth in the near future. Unfortunately, the Mars Sample Return mission will probably be delayed due to scheduling and budget issues.

"If AI can easily distinguish biotic from abiotic, as well as modern from ancient life, then what other insights might we gain? For example, could we tease out whether an ancient fossil cell had a nucleus, or was photosynthetic?," said Robert Hazen, co-author of the paper and an astrobiologist working at the Carnegie Institution of Science.

"Could it analyze charred remains and discriminate different kinds of wood from an archeological site? It's as if we are just dipping our toes in the water of a vast ocean of possibilities." ®

More about


Send us news

Other stories you might like