Particle.news
Download on the App Store

AI-Driven Chemical Analysis Finds Biosignatures in 3.3-Billion-Year-Old Rocks, Traces of Early Photosynthesis

A PNAS study uses machine learning on pyrolysis–GC–MS data to read degraded molecular echoes with over 90% accuracy.

Overview

  • An international team led by the Carnegie Institution for Science analyzed more than 400 diverse samples, from modern organisms to billion-year-old fossils and meteorites, to train and test the model.
  • The workflow identified biogenic molecular patterns in the 3.33-billion-year-old Josefsdal Chert in South Africa, providing the oldest chemical evidence of life reported to date.
  • Molecular signals of oxygen-producing photosynthesis were detected in rocks at least 2.5 billion years old, including South Africa’s Gamohaan Formation, with additional signals reported in Canada’s 2.30-billion-year-old Gowganda Formation.
  • The approach extends reliable molecular detection of ancient life from a previous limit of about 1.7 billion years and can distinguish categories such as photosynthetic versus non-photosynthetic organisms.
  • Researchers received NASA funding to adapt the method for planetary missions, and they are expanding the training dataset to improve performance on small or closely related sample classes.