Characterizing Biopatterns via Laser Ablation Mass Spectrometry

Abstract:

Since the Viking mission to Mars in the mid-1970s, the search of extraterrestrial life has focused on the origin, distribution, and evolution of carbon-rich-materials within the solar systems. The curiosity about whether life is universal has led to the deployment of subsequent space missions to comets, asteroids, rocky planets, and ocean worlds, while the quest about what is life has expanded the search of biomarkers beyond biology to broader contexts revolving around the coevolution of planets and life.

In addition to biomolecules, signatures of life could be present as biopatterns, recognizable morphological features (e.g., microfossils) and/or distinctive spatial distributions of substances (e.g., organics in inorganics matrix) that reflect biological influences in potentially habitable environments. Such repeatable morphological/chemical features could be related to microbe growth (e.g., cyanobacteria bind silica with biogenic carbonate), and/or biological responses to geological environments (e.g., coccoliths produce calcium carbonate plates to save energy in resource-limited conditions).

Laser Ablation Mass Spectrometry enables spatially-resolved measurements of organic (e.g., biomarkers) and inorganic (e.g., minerals) signals, and by extension 2-D chemical mapping of biopatterns without requiring physical contact with the planetary surface. However, the resolution and quality of chemical mapping increases logarithmically with the number of measurements, which imposes significant technical challenges to data volume, time, and energy requirements for spaceflight instruments.

This study proposes to evaluate the minimum number of measurements required to characterize the scale and magnitude of heterogeneity within a given 2-D sample by using kriging variogram models and machine learning. Kriging variogram models describe the spatial continuity of specified patterns and may be used to interpolate 2-D images from limited analyses based on model fitting to empirical data and residuals. We have created a database of kriging variogram models from >1000 different simulated patterns in MATLAB, and trained the weight of each model to real images by minimizing differences between predicted and real images via machine learning.