Researchers at the Boyce Thompson Institute and Cornell University have now developed a tool that begins to dismantle that bottleneck. The tool, called AIMe, short for AI Molecule Explorer, uses a form of artificial intelligence known as neuro-symbolic AI to predict, organize, and search the mass spectra of the entirety of known small organic molecules, a catalog of more than 100 million compounds. In doing so, it effectively builds a vast, searchable map of chemical space designed to accelerate both hypothesis generation and compound identification. The work is a collaboration between Frank Schroeder, Professor at the Boyce Thompson Institute and in Cornell’s Department of Chemistry and Chemical Biology, and Carla Gomes, Professor of Computing and Information Science and director of Cornell’s AI for Science Institute.

To understand why AIMe matters, it helps to understand how small molecule identification has worked for decades. Mass spectrometry is the workhorse technique, deployed everywhere from toxicology labs to food analysis facilities. When a compound is analyzed, the instrument fragments it and records the masses of the resulting pieces. That pattern of fragments, known as the tandem mass spectrum or MS2 spectrum, functions as a molecular fingerprint. To identify an unknown compound, researchers traditionally compare its spectrum against a reference library of spectra from known compounds, or they attempt to interpret the fragmentation pattern manually, developing hypotheses about the structure one painstaking step at a time.

The trouble is that experimental reference libraries remain sparse. Collectively, the available libraries cover fewer than 1 percent of known compounds, and resolving the structure of a single unknown can take days to months of iterative analysis and experimental validation. As a result, spectra without close library matches usually remain unannotated, no matter how abundant or biologically interesting they might be. Expert interpretation, meanwhile, does not scale. There are simply not enough specialists to work through the mountains of unannotated data accumulating in repositories around the world.

AIMe takes a fundamentally different approach. Rather than waiting for experimental spectra to accumulate in libraries, it predicts spectra computationally, then organizes those predictions into a searchable resource called MS2KOSMOS. Using this pipeline, the team generated more than 800 million predicted spectra covering essentially all known small organic molecules in PubChem, the largest publicly available chemical database. That represents roughly a thousandfold expansion of searchable chemical space relative to existing experimental libraries, a leap that transforms what is even possible to attempt when confronting an unknown spectrum.

At the heart of the system sits DeepMS2Reasoner, a model designed to simulate how molecules fragment inside a mass spectrometer. According to Gomes, the model builds fragmentation pathways step by step, using symbolic chemical rules to enumerate physically plausible fragmentation steps and a neural network to assign likelihoods to each step. The result is not merely a predicted spectrum but an annotated map of how a molecule came apart, a feature that makes AIMe’s outputs interpretable in chemical terms rather than just computationally useful. That interpretability matters enormously for working chemists, who need to understand why a prediction looks the way it does before they commit laboratory resources to testing it.

To demonstrate what the tool can accomplish in practice, the researchers applied it to a comparative metabolomics dataset from mice. The experiment compared germ-free mice, animals raised without any gut microbiota, against mice carrying a normal complement of gut bacteria. Several thousand chemical features differed between the two groups, and most of the abundant ones could not be identified using standard methods. The team used AIMe to query MS2KOSMOS with spectra from the 111 most abundant unidentified microbiota-dependent compounds. For roughly a third of them, AIMe retrieved close predicted spectral matches and related structural candidates, providing chemically interpretable leads for follow-up work. For the rest, the tool mapped the unknown spectra to molecular neighborhoods, sets of structurally related compounds whose shared fragmentation patterns could inform hypotheses about what the unknowns might be.

Two compounds in particular became a case study in what AI-guided structure elucidation can accomplish. Both produced spectra suggesting they were polyamine derivatives, a well-studied class of molecules, but their fragment patterns did not match anything previously described. Using AIMe’s output as a guide, Schroeder’s team assembled candidate structures combinatorially, predicted spectra for each candidate, and used the comparisons to narrow the field. For the first compound, the best candidate was a linear putrescine derivative, which was then easily verified by synthesizing an authentic standard. The second compound proved more elusive. The predicted spectra of potential candidates consistently failed to explain two prominent peaks, until the team expanded the search to include cyclized variants.

Synthesis ultimately confirmed what the predictions suggested. The second compound turned out to be a structurally unusual macrocyclic polyamine, a ring-shaped variant unlike any previously reported from mouse or human biology. Searches of a large public mass spectrometry database subsequently found the same compound in samples of human origin, detected in 57 of 99 human fecal samples examined. The discovery carries real biological weight. Polyamines occupy an important position in biology, sitting at the intersection of diet, the gut microbiome, and immune function, and these findings suggest that the catalog of microbiota-dependent polyamines is far less complete than scientists had assumed. A ring-shaped molecule that had gone unnoticed in human samples until now is a vivid reminder of how much chemistry remains hidden in plain sight.

The researchers also tested AIMe at repository scale, applying it to more than 7 million spectral clusters from the Global Natural Products Social Molecular Networking database, one of the largest publicly available repositories of mass spectrometry data. Prior annotation efforts had managed to annotate roughly 416,000 of those clusters. AIMe returned putative annotations for approximately 2.69 million, using the same similarity threshold applied in that earlier effort, a more than sixfold increase in coverage achieved computationally. AIMe is accessible through Cornell’s website, and the source code will be made publicly available upon publication of the study, with a preprint already posted on bioRxiv. If the tool performs as advertised across the wider community, the anonymous peaks that have long cluttered metabolomics datasets may finally begin to yield their secrets, and the hidden universe of small molecules that shapes human health may come into focus at a scale never before possible.

Subject of Research: Neuro-symbolic AI for predicting mass spectra and identifying unknown small molecules in metabolomics

Article Title: New AI tool maps the hidden universe of small molecules

Article References: New AI tool maps the hidden universe of small molecules. (n.d.). Original publication

Image Credits: AI Generated

DOI: Not provided

Keywords: AIMe, mass spectrometry, metabolomics, neuro-symbolic AI, small molecules, gut microbiome, polyamines, MS2KOSMOS, PubChem, compound identification, GNPS, chemical space

The experiment was elegantly simple in design. Young maize seedlings of the MV350 hybrid were grown hydroponically under controlled conditions, then exposed to an acclimating temperature of 15 degrees Celsius for three days. Crucially, one group experienced this cool period under normal growth light of 250 micromoles per square metre per second, while a second group was shaded at just 50 micromoles. Afterward, the plants faced either a severe cold challenge at 5 degrees Celsius or a salinity treatment with 0, 50, or 150 millimolar sodium chloride. The researchers then dissected the response at every level: chlorophyll fluorescence to track photosynthetic health, gene expression analysis by RT-qPCR, and an impressive metabolomics arsenal including GCxGC-TOF mass spectrometry for primary metabolites, untargeted LC-MS for conjugated compounds, and targeted profiling of hormones and polyamines.

The physiological measurements told a subtle story. The maximum quantum efficiency of Photosystem II, measured as Fv/Fm, barely declined after salt treatment, indicating that neither acclimation nor salinity inflicted irreversible damage. But the light-adapted quantum yield, Y(II), dropped significantly in plants exposed to 150 millimolar salt, and the decrease was more pronounced in plants that had previously been cold-acclimated. In parallel, the regulated non-photochemical quenching parameter Y(NPQ) rose, especially in cold-treated plants. Together, these signals suggest that moderate cold had sensitised the photosynthetic apparatus to subsequent osmotic stress, activating protective energy-dissipation mechanisms at the cost of photosynthetic efficiency, a classic trade-off seen in stress-primed plants.

The gene expression data revealed three striking patterns. First, the light-signalling genes ZmCOP1 and ZmHY5 were expressed predominantly in leaves and were strongly suppressed by low temperature, confirming that the light-sensing machinery itself responds to cold. Second, stress-responsive genes such as ZmERF33, ZmPMP3-1, and the sugar transporter ZmSWEET1a showed pronounced organ-specific regulation that depended on both temperature and light. Third, the polyamine oxidase gene ZmPAO1 remained comparatively active in stems even as it declined elsewhere, hinting that the stem is not merely a passive conduit but an active participant in coordinating the plant’s systemic stress response, possibly through redox signalling, since polyamine oxidases generate hydrogen peroxide as a by-product.

When salt was applied after cold acclimation, the transcriptional echoes of the earlier treatment were still audible. Most of the tested genes, including ZmPAO1, ZmPMP3-5, ZmZIP1, and ZmPIF1, were reduced under increasing salinity, but the magnitude and pattern of these changes depended on whether the plants had been light- or shade-acclimated. The most dramatic interaction appeared in ZmDREB80, a dehydration-responsive element-binding transcription factor. In shade-acclimated plants exposed to moderate salt, ZmDREB80 expression surged, particularly in stems, while in light-acclimated plants no such increase was detected. This single gene captured the essence of the study: the light conditions during a cold episode three days earlier had selectively reprogrammed how the plant would transcriptionally confront a completely different stressor.

The metabolomics results were equally compelling. Cold acclimation in the light triggered substantial accumulations of soluble sugars and amino acids in leaves, while the same cold period in shade produced far weaker changes. Under severe cold, sugars paradoxically decreased in light-acclimated leaves but increased in stems and roots, most likely reflecting reduced synthesis in the photosynthetic tissue combined with active export of photoassimilates to other organs. Amino acid pools behaved similarly, with root amino acid accumulation in the cold occurring only when acclimation had taken place in the light. The polyamine putrescine and its catabolite 1,3-diaminopropane rose significantly in leaves and stems, but only under light-acclimated conditions, while spermidine and spermine increased in the roots of light-acclimated plants under severe cold.

Hormone profiling added another layer of complexity. Abscisic acid, the central stress hormone, increased under both acclimating and severe cold, with the parallel rise in its degradation products confirming that the increase came from enhanced biosynthesis rather than slowed breakdown. Interestingly, the light conditions shifted where the abscisic acid accumulated: mainly in leaves and stems of light-acclimated plants, but in roots of shade-acclimated ones. Salicylic acid declined in leaves and roots during cold but rose in the stem under severe cold, while jasmonic acid and its active isoleucine conjugate dropped across all organs under severe cold, an unexpected response given that most studies report jasmonate accumulation during chilling.

Perhaps the most consequential finding emerged from the principal component analysis of primary metabolites in salt-treated plants. In every organ, the light-acclimated, salt-treated group separated clearly from the shade-acclimated and non-acclimated salt-treated groups, which clustered together. The loading vectors pointed to coordinated shifts in TCA-cycle intermediates such as citrate, malate, succinate, and fumarate, in soluble sugars, and in nitrogen-related metabolites including glutamate, asparagine, and citrulline. In leaves, photorespiration and redox compounds like glycine and serine dominated; in stems, phenylalanine and coumaric acid, precursors of the phenylpropanoid pathway, were decisive; in roots, asparagine and succinate pointed to nitrogen storage and respiration. This is the signature of a systemic, pathway-level reorganisation rather than a scatter of isolated metabolite changes.

The phenylpropanoid data reinforced the protective narrative. Chlorogenic acids increased in leaves of salt-treated plants in a concentration-dependent manner, more strongly in light-acclimated plants, consistent with their role as antioxidants and redox buffers. Ferulic acids rose more prominently in the roots of light-acclimated plants under high salinity, suggesting reinforced cell wall cross-linking at the soil-plant interface where ionic stress is first perceived. Feruloyl-putrescine conjugates, which combine polyamine and phenolic chemistry, accumulated almost exclusively in roots, hinting at a specialised strategy for wall reinforcement and ion flux regulation. Meanwhile, salt-induced accumulations of sugars, amino acids, and jasmonate signalling were actually dampened in light-acclimated plants, suggesting a shift toward energy-efficient osmotic regulation and reduced stress perception.

The authors are careful to note the limits of their interpretation. Because salt was applied immediately after the cold acclimation period, the design cannot distinguish a persistent acclimated physiological state from stress memory in the strict mechanistic sense, and the metabolic changes are indicators of altered physiological states rather than direct proof of mechanisms governing membrane stability or ion homeostasis. Still, the implications are significant. This is the first study to systematically map the organ-specific and systemic regulatory network underlying the light-temperature-salt interaction in maize, and it demonstrates that cold acclimation under sufficient light establishes a whole-plant, organ-coordinated state that reshapes subsequent salinity responses. As maize cultivation pushes northward into regions where sudden cold snaps and saline soils increasingly coexist, understanding how a plant’s environmental history primes its metabolism could become a powerful tool for breeders seeking resilient hybrids.

Subject of Research: Light-dependent cold acclimation and its effects on salt stress responses in maize

Article Title: Light-defined cold acclimation reprograms organ-specific metabolic and transcriptional landscapes for salt stress responses in maize

Article References: Benczúr, K., Gholizadeh, F., Pál, M., Janda, T., Hamow, K. Á., Dernovics, M., Nagy, K., Majláth, I., Darko, E., Wu, J., Van Le, N., & Szalai, G. (2026). Light-defined cold acclimation reprograms organ-specific metabolic and transcriptional landscapes for salt stress responses in maize. Plant Cell Reports, 45(10), Article 320. https://doi.org/10.1007/s00299-026-04008-z

Image Credits: AI Generated

DOI: 10.1007/s00299-026-04008-z

Keywords: maize, cold acclimation, salt stress, light signalling, metabolomics, gene expression, abscisic acid, polyamines, phenylpropanoids, photosynthesis, organ-specific response, stress priming