A machine learning program trained to differentiate between living and non-living things has a 90 percent success rate at differentiating the products of biology from those of non-living origin. Naturally, the makers could only train it on Earth-based life forms, so we can’t be certain it will be similarly successful when exposed to alien life, but hopes are high. Soon, they hope, we will have a chance to turn it loose on samples from Mars.
Some astrobiologists fear the Viking Lander found life on Mars, only to kill it by feeding it too much of a good thing. Even those who dispute this story worry we might not recognize extraterrestrial life if we do encounter it.
Professor Robert Hazen of the Carnegie Institution is preparing for that day, and has led a team who designed a program that they are teaching to differentiate life from non-life.
“We are asking a fundamental question; Is there something fundamentally different about the chemistry of life compared to the chemistry of the inanimate world?” Hazen said in a statement. “Are there “chemical rules of life” that influence the diversity and distribution of biomolecules? Can we deduce those rules and use them to guide our efforts to model life’s origins or to detect subtle signs of life on other worlds?”
Hazen and colleagues conducted pyrolysis gas-chromatography mass-spectrometry (GCMS) on 134 carbon-rich samples and fed the results to the program to see if it could find the patterns. GCMS involves heating materials without oxygen and then separating them so the component molecules can be identified by mass. Among the samples provided were 59 of biological origin, from shells and leaves to crude oil; the other 75 included samples from carbon-rich meteorites and even amino acids made in the lab.
The results have now been published, after initially being presented to a session at the Goldschmidt Geochemistry Conference. There, Hazen announced the program developed the capacity to distinguish the biological samples from the non-biological ones with better than 90 percent reliability. As with other forms of machine learning, the accuracy should improve further as more samples are presented.
Session co-chair Anastasia Yanchilina commented, “This moves us closer to recognising life when we find it.”
According to Hazen, “There are some interesting and deep implications which flow from this work. First, we can apply these methods to ancient samples from Earth and Mars, to find out if they were once alive. This is obviously important for looking at whether there was life on Mars, but it can also help us analyse very ancient samples from Earth, to help us understand when life first began.”
Impressively, the program even managed to exceed the challenge Hazen set for it. “We trained our machine-learning method on only two attributes - biotic or abiotic - but the method discovered 3 distinct populations - abiotic, living biotic, and fossil biotic – in other words, it could tell fossil samples from more recent biological samples,” Hazen said.
“This routine analytical method has the potential to revolutionize the search for extraterrestrial life and deepen our understanding of both the origin and chemistry of the earliest life on Earth,” Hazen said in another statement. “It opens the way to using smart sensors on robotic spacecraft, landers and rovers to search for signs of life before the samples return to Earth.”
Data already collected by Martian landers and rovers could reveal previously overlooked evidence for (possibly ancient) life.
The program appeared to learn that biological products achieve a level of complexity that is rare to non-existent without life. Complex molecules that were useful turn up frequently in any product of life. The specific molecules vary depending on the life that made the product, and what it was for, but the pattern of frequency remains.
Professor Emmanuelle Javaux of the University of Liège, who was not directly involved in the work, commented, “It would also be very interesting to test this new method on some of the oldest putative and debated traces of Earth life.”
Nevertheless, the problem remains that all the products of life used in the study are branches of the same metaphorical tree. We cannot know how similar or different life would be if it evolved in isolation from an entirely different source. Hazen is optimistic, saying, “[Our] method should be able to detect alien biochemistries [...] [by looking] for patterns in molecular distributions that arise from life’s demand for ‘functional’ molecules.”
The study is published in Proceedings of the National Academy of Sciences.