Research preview

Map any MS2 spectrum across the known structural space

AI Molecule Explorer combines spectral prediction, molecular search, and interpretable fragmentation to contextualize experimental spectra across MS2KOSMOS.

Paste m/z–intensity pairs, import an .MGF or .MSP file, or begin with a preloaded example.

MS2KOSMOS

~105Mmolecules
800M+predicted spectra
4collision energies
2ionization modes

A multi-agent framework

Three specialized agents

AIMe enables query-driven molecular contextualization across the Known Organic Small-Molecule Space through a modular neuro-symbolic workflow.

Deep­MS2­Reasoner

Predicts MS2 spectra and interpretable fragmentation pathways at experimental collision energies.

Predict + explain

MS2­KOSMOS­Generator

Constructs and indexes a searchable space of predicted spectra for known small organic molecules.

Construct + index

MS2­KOSMOS­Mapper

Maps query spectra onto MS2KOSMOS for molecular search, comparison, and visualization.

Search + map

Search MS2KOSMOS

Collision energy–matched search of the MS2KOSMOS

Match experimental spectra against predicted spectra of >105 million molecules at NCE 20%, 30%, 50%, and 80%, in both [M+H]+ and [M-H]- ionization modes.

Start spectrum search
CE Recommended

Collision energy—matched spectra

Retrieves candidates, then predicts and reranks their spectra at the experimental collision energy in NCE or eV.

✓ Fine-grained spectral prediction ✓ Detailed fragmentation DAGs
PC

Precomputed spectra

Searches spectra calculated at the four indexed collision energies for a faster, coarse-grained result.

✓ Faster candidate search ✗ No detailed fragmentation information

Human-in-the-loop

Hypothesize–predict–compare AI-assisted loop

For any unknown spectrum, AIMe enables retrieval of potentially related molecules within MS2KOSMOS. Retrieval works over shared fragment formulas, and thus stays informative even if the true structure is not in PubChem, in that substructures of the retrieved candidates can serve as foundations for developing structural hypotheses. Retrieved candidates thus become priors for AI-assisted structure elucidation.

Hypothesize

Propose one or more possible structures, e.g., based on initial search results.

Predict

DeepMS2Reasoner predicts spectra and fragmentation DAGs for proposed structures.

Compare

Predicted spectra are compared and ranked against the experimental spectrum, enabling interpretation and refinement of structural hypotheses.

Each comparison guides the next hypothesis: an iterative, interpretable path to structure elucidation, with the scientist in control.

Try on search page
Experimental ↔ predicted MS2
Mirror plot comparing an experimental MS2 spectrum with its predicted spectrum
Fragmentation DAG
A Fragmentation DAG with probability threshold and graph layout controls

Explore the result

Detailed fragmentation pathways for every spectrum

Inspect candidate molecules, mirror plots, matched fragments, and Fragmentation DAGs. Open results directly after a search or reload a saved .json file at any time.

Research preview

Built for research, exploration, and discovery

For more details, see the accompanying paper:

Utku U. Acikalin, Dieqiao Feng, Aaron M. Ferber, Goncalo J. Gouveia, Tyler J. Schwertfeger, Di Chen, Delia Qu, Marissa A. Fontaine, Yingheng Wang, Richard A. Bernstein, Haofan Wang, Tae-Hyung Won, Christopher N. Parkhurst, Bart Selman, David Artis, Frank C. Schroeder, Carla P. Gomes. Charting the small-molecule universe from mass spectra with neuro-symbolic AI. bioRxiv (2026) doi:10.64898/2026.08.05.743095.

We thank the project's sponsors for their support: Schmidt Sciences, National Institutes of Health (NIH), National Institute of Food and Agriculture (USDA/NIFA), Air Force Office of Scientific Research (AFOSR), Biocodex Microbiota Foundation, and Brain and Behavior Research Foundation.