In the average experiment that uses mass spectrometry (MS) to characterize a biological sample’s small-molecule metabolites, scientists are unable to define the chemical structure of 90% of the molecules they can detect. Methods for MS have evolved, but the molecules that make up what is called the dark metabolome continue to stymie researchers.
Metabolomics has a data problem. Enveda, a biotechnology company focused on harnessing the biological properties of natural products, is taking a crack at this problem by relaunching the Critical Assessment of Small Molecule Identification (CASMI) contest. CASMI ran sporadically from 2012 to 2022, but Enveda says it hopes the appeal of the new contest will extend beyond the MS metabolomics community to the broader machine learning crowd.
The aim is to create an algorithm that can correctly identify a molecule’s structure based on its tandem MS spectra—the fingerprint of fragments that a molecule breaks into after being ionized.
“We think there’s a line of sight to be able to characterize the world’s chemistry at scale, essentially akin to what sequencing did to DNA,” says Enveda CEO Viswa Colluru.
CASMI owes a lot to the Critical Assessment of protein Structure Prediction (CASP) competition, which famously catalyzed the breakthrough protein-folding artificial intelligence model AlphaFold. “CASMI was inspired by CASP,” says August Allen, chief technology officer at Enveda. “The difference is CASMI hasn’t yet had its Alphafold moment . . . and that’s what we’re trying to usher in.”
But while CASP participants could lean on nearly 50 years’ worth of structured data collected in the Protein Data Bank, the data available for metabolomics are messier. And getting good training data is important: without information about ground truths, algorithms can’t make predictions.
As part of the 2026 CASMI, Enveda is providing a training dataset of MS/MS spectra from its own collection of purified natural products. The company has also standardized a collection of open-source metabolomics datasets that competitors can use for training. According to Allen, that may entice computer scientists who don’t have a deep metabolomics background.
To test the models that are entered, Enveda has curated a test set with spectra from molecules that have not been characterized in public datasets.
Oliver Fiehn, a computational chemist at the University of California, Davis, ran the 2022 iteration of CASMI. He says that the dataset that entries will be tested on is less complex than a genuine metabolome. “They did not try to have real datasets with . . . overlapping coeluting entities, chimeric spectra, weird adducts,” Fiehn says. “They focused on mass spectra of isolated compounds, which is difficult enough.”
Fiehn says he sees excitement among early participants in the competition but doesn’t expect a breakthrough this time around. “AlphaFold took CASP many years,” he says, adding that in metabolomics there is “limited information and much less ground truth, in a way.”
Past analyses of using machine learning for MS annotation have concluded that models struggle to recognize the same molecule captured under different experimental conditions and to extrapolate beyond the structures in their training datasets; the available training data, according to one analysis, do not offer broad coverage of known metabolites.
In an email to C&EN, Allen counters that these are exactly the problems CASMI is designed to solve.
The competition is open until mid-December and offers a $50,000 prize to be split between the top competitors.
