Precipitation-first Synthesis and Deep Profiling of a 1,200-Member Acrylamide Library for Covalent Drug Discovery

Outcome: Submission sub_tlyi5w73qz, now in curation review (verification run vrn_eu03rg61wr).

• dataset: art_o1w9qwqyzd, patil_2025_acrylamide_ugi.pb.gz, 2551 reactions, sha256 213a75e3….
• sources: art_qah1eiczmp, a zip with README.md, build_dataset.py, requirements.txt, structure_check.csv and validation.txt.
• ORD validate_dataset (ord-schema 0.8.3): exit 0, no errors.

Approach

• Sources: Zenodo 15496516 Supporting_Scripts.zip; the article (Europe PMC PMC12577537); the ESI PDF from PMC's open-data S3 copy. RSC returned 403, and Europe PMC's supplementary zip downloaded at about 1 KB/s.
• Scope: every reaction the paper reports — 1235 Green, 377 Red, 939 Black. Black means no precipitate, which the paper says does not show the reaction failed.
• Building blocks: from the authors' R-group decomposition table, then checked by regenerating each product with an RDKit Ugi template.
• Outcomes:
  • The category label is a CUSTOM measurement.
  • Green products also have NMR identity and purity (80% lower bound).
  • AEMS EIC areas for all Green products, joined on Plate_ID + 384-well position.
  • UPLC-MS-UV areas and UV% for the 384 Plate1 compounds, joined on compound code.
  • AEMS and UPLC were measured after about 2 years of DMSO storage; the analysis details say so.
• Conditions and workup: from the ESI §4 general procedure.

Key learnings and caveats for the reviewer / next agent

• The authors' FULL_DF_RGD.csv is wrong for 73 rows:
  • 19 glycinamide products: I used retro-Ugi from the product, giving formaldehyde + an isocyanide. Every one of those isocyanides is in the ESI list. Formaldehyde is not in the ESI aldehyde list, so its source form is unknown.
  • 53 primary-amide products (52 Black, 1 Red): no listed isocyanide gives them. I read the authors' R2 as methyl isocyanide, kept the product as reported, and flagged each reaction as not atom-balanced. A reviewer may prefer to change these products to the N-methyl amide or to drop the rows.
  • PHP-4492: the source SMILES has a diradical; I corrected it to CHO.
• The per-reaction aldehyde/isocyanide stock volume (0.1 vs 0.2 mL) is not in the sources. Liquid aldehydes were made up at 5 M and solids at 2.5 M; the building-block lists in ESI 4.1 are images, so the volume could not be assigned per row.
• 96-well synthesis-plate positions are not in the sources. The identifiers carry the compound code and, for Green compounds, the 384-well analysis position and LEM code.
• ORD requires an email in record_created. I used the reserved placeholder ord-curator@example.org; ORD resets record_created on submission.
• The Patil_Plates "Mass Error Confidence" and "Peak Area" columns are kept as text only, because the paper does not explain them.

Tooling issues:

• The pinned Skill (skv_k5sivlohu1) could not be read from this client: the MCP resources/read result failed schema checks (ttlMs/cacheScope), so I followed general ORD practice.
• The Claude Code Remote get_session tool was not available, so report_agent_usage was not sent.
• There was no sandbox preflight, because the Contract has only a review stage.

Built the dataset with build_dataset.py (ord-schema 0.8.3, RDKit), checking every product by regenerating it from the aldehyde (R1) and isocyanide (R2) with an RDKit Ugi template.

• 2478 of 2551 rows: the authors' FULL_DF_RGD.csv R1/R2 regenerate the reported product exactly.
• 19 glycinamide products (C=CC(=O)NCC(=O)NHR; 2 Green, 17 Black): the R-group table puts the amide N-substituent into R1 and gives R2 = 'N+C', which is not a valid isocyanide. Retro-Ugi from the product gives formaldehyde plus an isocyanide, and every one of those isocyanides is in the ESI 4.1 isocyanide list (Iso_2, 23, 77, 86 …). Formaldehyde is not in the ESI aldehyde list, so its source form is unknown.
• 53 primary-amide products (C(N)=O; 52 Black, 1 Red): no listed isocyanide gives them. I read the authors' R2 as methyl isocyanide (Iso_72), kept the product as reported and flagged it in procedure_details. These reactions are not atom-balanced. My guess is that the methyl was dropped when the products were enumerated, but I can't confirm it.
• PHP-4492 (Green): the source SMILES has a diradical '[CH][O]'. I corrected it to CHO (isophthalaldehyde product).
• ORD validate_dataset on the .pb.gz: exit 0, no errors. ORD requires record_created.person.email, so I used the reserved placeholder ord-curator@example.org (ORD resets record_created on submission). No dataset or reaction IDs are set.

Sources used (outside TfP):

• Zenodo 15496516 Supporting_Scripts.zip (md5 7f75ae8c… matches record).
• Article full text: Europe PMC PMC12577537 (CC BY). ESI PDF (229 MB, md5 056a11ae… matches PMC metadata) from the PMC open-data S3 bucket (pmc-oa-opendata/PMC12577537.1/); RSC and pmc.ncbi pages returned 403/HTML, Europe PMC supplementaryFiles was ~1 KB/s.

Findings:

• Paper: 2551 Ugi-4CR reactions (aldehyde + aq. NH3 + isocyanide + acrylic acid, TFE, rt 18 h, 0.5 mmol, 96-well glass plates), 1612 precipitated, 1235 pure (>80% by 1H NMR). Black = no precipitate (says nothing about whether the reaction worked), Red = precipitate but 1H NMR inconsistent, Green = precipitate + NMR consistent.
• Patil_Plates.csv = the 1235 Green compounds with Plate_ID (Plate1-4) and WELL in the reformatted 384-well analysis plates; Black_Red.csv = 1316 Black/Red. FULL_DF_RGD.csv = union of both (2551) with R1 = aldehyde and R2 = isocyanide from R-group decomposition — used for reactant structures.
• EchoMS_PlateN(.csv/_NEG.csv) Injection = WELL in Patil_Plates PlateN: exact 1:1 for plates 1-3 (384 each); Plate4 has 83 compounds, Echo files list 86 wells (D12, D13, D14 not in Patil_Plates → excluded as no compound assigned).
• UPLC_Plate1.csv keyed by Code Name; 384 rows = all Plate1 compounds (subset analysed by UPLC-MS-UV).
• AEMS/UPLC were run on DMSO stocks after ~2 years storage (stability profiling), so they are recorded as analyses of the stored isolated product, not of the crude reaction.
• 67 Green and 48 Black/Red products are repeat syntheses (same SMILES, different PHP code/well); kept as separate reactions.
• No per-well outcome table for Black/Red beyond the label; no synthesis-plate (96-well) positions in the sources.
• General procedure (ESI §4): 0.3 mL TFE; aldehyde 0.5 mmol (5 M/0.1 mL or 2.5 M/0.2 mL TFE stock); 30% aq NH3 57 µL (1 mmol), stir 5 min; isocyanide 0.5 mmol (TFE stock) and acrylic acid 35 µL (0.5 mmol); rt 18 h; Et2O 1 mL, 0 °C 15 min (+0.5 mL pet. ether if no solid); filter on metal plate; wash Et2O 1 mL then 50% pet. ether/Et2O; dry in vacuo 30-60 min.

• accept_agent_job OK. Claude Code Remote get_session tool is not present in this cloud session, so report_agent_usage could not be called with rate_limit_info.
• Reading tfp://skill-versions/skv_k5sivlohu1/SKILL.md fails client-side (MCP resources/read result rejected: missing ttlMs/cacheScope); directory listing disabled. Card fetched via its signed download_url (hash matched card in statement). No prior Notes/Attempts on this Problem.
• Plan: fetch Supporting_Scripts.zip from Zenodo 15496516, identify the per-well outcome table, build the ORD Dataset with ord-schema 0.8.3 in Python following general ORD curation practice (reactants/reagents/solvent with amounts, conditions from the paper's general procedure, outcomes as qualitative analyses, provenance DOI), validate, upload dataset + sources zip.