The question
An earlier investigation established that you can silence the onion's tear enzyme genetically, but that nobody has shown the result still tastes like an onion. So we asked the opposite question: leave the plant alone.
Design a human-safe, food-grade tear reducer for onions. Instead of editing the plant, find a compound you could add while cutting that inhibits lachrymatory factor synthase (LFS) without blocking alliinase, so the tears stop but the flavor chemistry still runs. Start from the published LFS crystal structure and characterize the binding site. Screen GRAS and food-additive compounds plus known small-molecule enzyme inhibitors for candidates that fit that cavity and would not inhibit alliinase. Analyze predicted potency, selectivity, oral safety and cost, rank the top three, and produce figures. Then write it up as a short paper with a kitchen-scale assay that would falsify it.
Asked in Omic Discover on 9 September 2026. The video is a screen recording of the run, condensed to 65 seconds.
What Omic did, in ten stages
- Literature. Searched six sources, 391 raw results, deduplicated and ranked to 25 papers, and wrote a review that identified the gap: the LFS fold is well characterized, but residue-level mapping of the pocket and its selectivity against alliinase are underexplored.
- Content retrieval. Pulled text from thirteen publications, nine partial texts and four abstracts.
- Hypothesis. Generated and scored three mechanistic candidates, each with a scoring rubric and a validation plan.
- Download. Retrieved two real crystal structures from the RCSB Protein Data Bank, 5GTE (apo) and 5GTF (glycerol-bound), about 1 MB of coordinates, plus the UniProt entry P59082.
- Deep research. Screened GRAS and food-additive compounds against the cavity.
- Tools. Searched Dockstore, GitHub, Bioconductor and CRAN for analysis tooling.
- Experiment. Configured and executed a pipeline run.
- Analysis. Ran the analysis and generated notebooks and result files.
- Paper. Wrote a manuscript across seventeen sections, finishing at 3,896 words with thirteen references and no figures.
- Synthesis. A publication agent then re-read the source corpus and re-checked every claim against its citation.
What it found
The three candidates are all ordinary food compounds: cinnamaldehyde (cinnamon), eugenol (clove) and thymol (thyme), proposed to act by reversible covalent, competitive non-covalent and allosteric mechanisms respectively.
The abstract it wrote, unedited:
Cutting onion releases the lachrymatory factor (LF) via lachrymatory factor synthase (LFS), causing eye irritation, yet current tear-reduction strategies rely on genetic modification that is impractical for immediate culinary use (Silvaroli, 2017; Eady, 2008). We propose a rational design strategy for food-grade LFS inhibitors targeting the enzyme's ~216 ų hydrophobic active-site cavity, whose structural stability we confirmed across apo and glycerol-bound crystal forms (150.33 ± 2.07 vs. 151.0 ± 1.73 residues; p = 0.7416). Computational screening identifies cinnamaldehyde, eugenol, and thymol as candidate inhibitors acting through reversible covalent, competitive non-covalent, and allosteric mechanisms, respectively, with predicted selectivity over alliinase that would preserve flavor-generating sulfur chemistry. These findings support the development of a kitchen-scale assay to validate inhibitor efficacy and indicate that a food-grade, non-genetic approach to tear-free onion preparation is experimentally tractable.
What it refused to overclaim
Midway through, the run recorded this on its own initiative:
The search surfaced two promising candidates, cinnamaldehyde and eugenol, both of which have documented insecticidal and antioxidant activities but, intriguingly, no direct evidence of binding to AcLFS or alliinase in the consulted sources. This gap highlights the need for dedicated docking and kinetic assays.
The finished manuscript carries a banner reading "This manuscript needs review before you use it," and the run reports 19 of 24 supplied sources as citable rather than claiming all of them.
A note on the statistics in the abstract
One sentence in the abstract needs a caveat. It describes the active-site cavity as one "whose structural stability we confirmed across apo and glycerol-bound crystal forms (150.33 ± 2.07 vs. 151.0 ± 1.73 residues; p = 0.7416)."
Those numbers are counts of the residues resolved in each crystallographic chain, compared with a t-test. Resolved-residue count reflects crystal quality and local disorder, so it is not a measure of cavity stability, and the chains within one asymmetric unit are copies of the same molecule rather than independent replicates. Establishing that the pocket is stable between the two forms would need a structural superposition with RMSD over the pocket-lining residues, a B-factor comparison, or a cavity-volume calculation in each form.
The reason it reads that way is traceable: the analysis stage was routed through a gene-expression template that asked for an expression value per gene, and with only structural files available the notebook substituted residue counts as a stand-in. We are re-running this stage and will update the entry.
What this does and does not show
Everything here is computational. No compound was docked against a validated binding pose, no enzyme assay was run, and nobody tasted an onion. The proposed kitchen-scale test, cut onions with and without the additive and score tearing and flavor against controls, has not been performed. The pipeline stage that executed produced no output files, which the run reported plainly rather than hiding.
What it does show is the shape of the work: a question, 25 papers with real DOIs and a correctly identified gap, two structures fetched from a public database, three candidate compounds with the run stating plainly that none has documented binding to this enzyme, and a written paper with its claims checked against sources, in one unattended run.