A MACHINE LEARNING APPROACH FOR THE DETECTION OF AGNOSTIC MOLECULAR BIOSIGNATURES

Is the diversity and distribution of molecules in living systems different than those in nonliving systems? Can we find a method to characterize the organic molecules in order to understand the differences in the synthesis pathways in biotic versus abiotic processes? We propose that there exist “agnostic molecular biosignatures” composed of suits of molecules that differentiate terrestrial biochemistry from abiotic chemistry.

A diverse collection of natural and synthetic organic molecular mixtures was examined using pyrolysis gas chromatography-mass spectrometry (py-GC-MS), which decomposes the samples into fragment ions for molecular identification. We trained machine-learning models using three-dimensional chromatographic retention time/mass-to-charge ratio/intensity data from each sample which resulted in a model that can differentiate between biotic or abiotic samples with ~90% accuracy. The relational characteristics of chromatographic and mass to charge ratio provide the needed information to differentiate between the biotic and abiotic groups and hence, the method does not need the exact compound identification.

In this talk we present preliminary results that suggest that organic pyrolysis products cluster into three groups: organic material derived from abiotic sources (material from laboratory prebiotic chemistry simulations, carbonaceous meteorites etc.), living terrestrial matter (living cells, microbes, plants, etc.), and geologically processed biotic organics (including coal, oil shale, petroleum, etc.). We discuss the agnostic nature of our proposed biosignature and how it will likely be useful for the detection of alien biology as even alien biochemistry might be interpretable and differentiable from the chemistry from abiotic processes.