Particle.news
Download on the App Store

AI-Enabled Chemical Analysis Finds 3.3-Billion-Year-Old Biosignatures

A peer-reviewed PNAS study unveils a Py-GC-MS machine-learning workflow that reads degraded molecular patterns to extend the reliable chemical record of ancient life.

Overview

  • Researchers report molecular evidence of microbial life in 3.33-billion-year-old South African rocks and photosynthetic activity in samples at least 2.5 billion years old.
  • The approach pairs pyrolysis–gas chromatography–mass spectrometry with supervised machine learning to classify biotic versus abiotic chemistry with greater than 90% accuracy.
  • The model was trained and tested on 406 diverse samples spanning modern organisms, ancient fossils, sediments, meteorites, and synthetic mixtures.
  • The work roughly doubles the age range for trustworthy molecular biosignatures from about 1.6–1.7 billion years to at least 3.3 billion years and pushes the chemical record of photosynthesis back by more than 800 million years.
  • The team has NASA funding to develop flight-capable instrumentation, while seeking many more and more diverse samples after noting lower accuracy for closely related classes and small training sets.