Enveda launches a competition to determine molecular structures from mass spectra

Enveda launches a competition to determine molecular structures from mass spectra

On 14 September, biotechnology company Enveda launched CASMI 2026 on Kaggle, a competition to determine the structures of small molecules. Participants receive tandem mass spectra, which record the fragments produced when a molecule breaks apart in the instrument, and propose its two-dimensional structure. The prize pool is $50 thousand; submissions close on 14 December.

A mass spectrometer converts a molecule into an ion, breaks it into fragments, and measures their mass-to-charge ratios. A chemist or algorithm uses these signals to work out how the atoms were connected. Researchers encounter many such signals when analysing a biological sample. The structures of approximately 90% of small molecules remain unknown, according to C&EN. These unidentified compounds are known as the dark metabolome.

CASMI began in 2012 to compare methods for identifying small molecules using a shared dataset. Organisers confirm each sample’s structure in advance, then give participants its spectra. In the paper on the first CASMI, the competition’s founders explained why selecting suitable samples is difficult: their structures must be confirmed, while a library search should provide only part of the solution.

Enveda has applied this approach to an open competition for machine learning specialists. The company supplied spectra of purified natural compounds for training and standardised public datasets into a common format. The hidden test set contains spectra of molecules whose structures were absent from public datasets.

Under the CASMI 2026 rules, an algorithm proposes up to 25 candidate two-dimensional structures for each spectrum. The MRR@25 score increases as the first correct structure appears closer to the top of the list. The hidden test set keeps the answers with the organisers until evaluation, allowing algorithms to be compared on the same task.

Oliver Fiehn, a computational chemist at the University of California, Davis, who ran CASMI in 2022, described the selected dataset as follows:

“They focused on mass spectra of isolated compounds, which is difficult enough in itself.”

A biological sample contains several substances at once; their signals can overlap, and additional ions can attach to the molecules. The competition uses purified compounds, so it measures how accurately an algorithm matches a single spectrum to a single structure.