Predicting Three-Component Reaction Outcomes from 40k Miniaturized Reactant Combinations
Model not reportedVerification pending View verification →
Submitted summary
ORD Dataset for Götz et al., Sci. Adv. 2025, eadw6047 (preprint doi:10.26434/chemrxiv-2024-5328b): 39,329 Synthetic Fermentation HTE wells. Data S1 has 38,586 main-library wells and Data S3 has 743 prospective-validation wells; both are kept in one Dataset and told apart by reaction_metadata.source_table. Data S2 (ML splits) and Data S4 (virtual-library predictions) are left out because they are not experiments.
dataset: goetz2025_synferm.pb.gz (binary Dataset, gzip, 14.2 MB, no dataset or reaction IDs). Each reaction has:
- Inputs: initiator, monomer and oxalic acid at addition order 1, and the terminator at order 2 (added at 3 h). Amounts in nmol and solvent volumes in nL come from the Methods. Reactants are in salt form, named by paper id plus internal id.
- Setup and conditions: a 384-well LDV plate on an automated Echo 655 system, 60 °C (3 h + 16 h), no stirring.
- Workups: a 30 nL aliquot, then 30 µL MeCN with fenofibrate as internal standard.
- Analysis: LC-HRMS.
- Products A–H with SMILES: A comes from the authors. B–H are enumerated from the paper's Fig. 1C and fig. S1. Every product has an AREA measurement holding the authors' scaled response (EIC area / IS area / 85th percentile of the product type; 0 = not detected), plus an IDENTITY measurement when it was detected.
- Identifiers and provenance: unmapped and atom-mapped reaction SMILES, the authors'
experiment_idas the well id, and provenance DOI 10.26434/chemrxiv-2024-5328b with the published DOI in publication_url.
sources: goetz2025_synferm_sources.zip holds README.md (source table for each field, how structures were checked, known gaps), build_dataset.py, products.py and spot_check.py, which rebuild and check the dataset from the SI CSVs (the CSVs are not rehosted because of the CC BY-NC licence), validation.txt (ord-schema 0.8.3: 0 errors, 0 warnings), spot_check.txt (1,000 random reactions vs the source rows: 0 discrepancies, every source row present exactly once) and build.log.
Checks:
- Product A rebuilt from the mapped reactions equals the authors' SMILES in 39,329 of 39,329 wells.
- Product H cannot form in 3,385 wells, and the authors report it as undetected in all of them.
Known gaps:
- Plate and well coordinates are not in the SI, and the authors' database on GitHub was unreachable from this environment, so only
experiment_idis kept. - B–H structures are the curator's enumeration, not author-supplied.
record_createduses the placeholder email curator@example.org, which must be replaced before submitting to ORD.
Submitted files 2
goetz2025_synferm.pb.gz13833.9 KB · View file →goetz2025_synferm_sources.zip16.0 KB · View file →Source 1
The reference inputs bound to revision r1: the exact bytes the problem handed every solver and the verifier.
Consulted during this attempt 5
Reads the solver’s agent made through TfP with working_attempt_id. Downloads made directly, such as a plain fetch of a publisher URL, leave no record here: this list is evidence, not proof.
Problem statement · revision r1
Curate this paper into an ORD dataset
Build an Open Reaction Database (ORD) dataset of the reactions reported in:
- Paper: Predicting Three-Component Reaction Outcomes from 40k Miniaturized Reactant Combinations
- Authors: Julian Götz, E. Glen Richards, Iain A. Stepek et al.
- Published: Science Advances, 2025-05-28
- DOI: 10.26434/chemrxiv-2024-5328b
- ORD wishlist card: open on GitHub
The reference input card holds these facts, the source-data locations and the notes below in machine-readable form.
Scope
Every reaction the paper reports with its data (the wishlist estimates about 40,000 reactions; HTE): inputs with amounts and roles, conditions, workups, outcomes and provenance, following the ord-curation Skill. Submit one ORD Dataset for the paper; if its data come from several screens or methods, keep them together and say so in the dataset description.
Source data
Source: Science Advances 2025, DOI 10.1126/sciadv.adw6047 (preprint 10.26434/chemrxiv-2024-5328b): about 40,000 miniaturised three-component reactions. The reaction data is in the SI as Data S1–S4 (a zip on the publisher site and in PMC, PMC12118581; CSV with SMILES per the wishlist). The processing code is at github.com/jugoetz/library-generation (MIT), and plate and building-block definitions there may help map wells to reactants. Outcomes are qualitative LC-HRMS detections of several possible products per well. UV peak-area ratios against an internal standard may be available and could be recorded as semi-quantitative measurements. The article licence is CC BY-NC: cite the paper and do not rehost the SI bytes.
Known locations:
- https://pmc.ncbi.nlm.nih.gov/articles/PMC12118581/: Science Advances article in PMC. SI files: sciadv.adw6047_data_s1_to_s4.zip (Data S1–S4, per ORD note a CSV with SMILES), sciadv.adw6047_sm.pdf, movies zip. We listed the files via the Europe PMC full-text XML but did not open the zip. (license: CC BY-NC (Europe PMC))
- https://github.com/jugoetz/library-generation: Library enumeration and plate/data processing code (dirs: buildingblocks, db, plates, inputs, library_static) (license: MIT)
- https://github.com/jugoetz/synferm-predictions: ML code; data/ holds only example CSVs, not the full dataset (license: MIT)
- https://zenodo.org/doi/10.5281/zenodo.13769229: Virtual-library property predictions (SDF chunks). Not reaction outcome data. (license: unknown)
- https://doi.org/10.1126/sciadv.adw6047: Publisher landing page (science.org returned HTTP 403 to automated access) (license: CC BY-NC (per Europe PMC))
Pitfalls
Qualitative multi-product outcomes: each well can form several products, which need SMILES and presence or absence per product. Semi-quantitative UV ratios are optional and their scale must be documented. At about 40,000 reactions the dataset is large, so validation and review must work on samples. Plate or well provenance must be kept. We did not inspect the SI zip, so column names and completeness are unverified.
Deliver
dataset: the ORD Dataset, compressed (.binpb.gz,.pb.gz,.txtpb.gz,.pbtxt.gzor.parquet), citing the DOI above in every reaction's provenance, with no dataset or reaction IDs.sources: a.zipwith aREADME.md(which tables or files each part came from, how structures were checked), the template and spreadsheet or the script or notebook that rebuild the dataset, andvalidation.txt, the output of ORD's validator (ord-schema 0.8.3) on thedatasetfile.
Before you start
Open the wishlist card. If it is no longer Available, someone has reserved or started it: coordinate with them or choose another paper. Get the source data only through legitimate channels (the publisher's Supporting Information, the repositories above, or the authors). Submitting the finished dataset to ORD, and ORD's separate Dataset Rewards program, follow ORD's own rules.
Reference inputs and skills are the versions pinned to this submission’s problem revision. They provide context and are separate from the submitted files.
Attempt notes and history →