Seminal
The magazine and atlas of EU-funded researchA new publication from the web and AI team at Scott Charnley

Atlas / Spain

Marie Skłodowska-Curie Action · Running

FoundaMet FoundaMet: A Foundational Model for Metabolomics

Coordinator
Fundacion Universitaria San Pablo-Ceu, Spain
Runs
1 February 2027 to 31 January 2029
Status
Running
EU funding
€194k
Total cost
—
Partners
1 organisations in 1 countries
Type
Marie Skłodowska-Curie Action
Grant agreement
101281191

What the project is doing

In the project's own words, from its summary on CORDIS.

Untargeted metabolomics is central to biology and medicine, yet a large fraction of mass-spectrometric (MS) experimental signals remains unannotated. Current identification hinges on comparing experimental properties—retention time (RT) or migration time (MT), collision cross-section (CCS), and fragmentation spectra (MS/MS)—with reference libraries. These libraries are incomplete, costly to expand, and often instrument- and method-specific.

Machine learning can mitigate this gap by predicting such properties directly from the compounds chemical structure, thereby enabling identification beyond existing libraries. However, existing models frequently generalise poorly across datasets and laboratories because available training sets are too small to support robust, transferable deep learning. This project will develop FoundaMet, a foundation model for metabolomics that enables reliable knowledge transfer for molecular-property prediction.

FoundaMet will learn chemical structure–aware embeddings from molecular graphs via large-scale self-supervised pre-training to capture local bonding patterns and long-range substructures, followed by supervised multi-task learning that simultaneously predicts multiple molecular properties. The resulting model will provide general reasoning capabilities over metabolites, improving cross-instrument and cross-laboratory generalisation and achieving higher accuracy on task-specific predictions even when only small annotated datasets are available.

On top of the pre-trained backbone, we will build task-specific heads for predicting RT, MT, CCS, fragmentation spectra (MS/MS), and adduct formation. This functionality will be integrated into the CEU Mass Mediator metabolite-annotation platform to facilitate immediate uptake by the metabolomics community.

Who is involved

OrganisationCityCountryRole
Fundacion Universitaria San Pablo-CeuMadridSpainCoordinator

Source: CORDIS, European Commission. Data of 5 Nov 2026, reused under the Commission's reuse policy. View on CORDIS · Grant DOI