{"success":true,"data":{"author":"OpenADMET and Schrodinger","bucketAccessType":"public","bucketName":"openadmet/pka-modeling-data","bucketPrefix":"","category":"Drug Discovery","cloudProvider":"huggingface","createdAt":"2026-11-22T01:09:20.979Z","description":"Experimentally measured aqueous acid dissociation constants (pKa) for 717 small molecules from the OpenADMET and Schrodinger collaboration, measured on a Pion SiriusT3. The compound set deliberately includes chemistry that is hard to predict: tautomerizable compounds, electron-deficient heteroaromatics and heterocycles with little literature precedent, alongside carboxylic acids, phenols and substituted pyridines. Most compounds were screened by UV-metric titration (UV-vis absorbance as a function of pH) and lower-confidence ones confirmed by potentiometric pH-metric titration, with three titrations at different methanol cosolvent ratios extrapolated to 0% cosolvent by Yasuda-Shedlovsky extrapolation over pH 2 to 12. Every measurement was fitted with 1-, 2-, 3- and 4-pKa models in SiriusT3Refine and all fits are reported, so the single CSV (7,246 rows) gives Enamine ID, SMILES, assay method, pKa model, pKa value and standard deviation, ionic strength, temperature, cosolvent fractions, manual QC flag and comment codes; users choose a model per compound before training. Use it for pKa and protonation-state prediction and ADMET modeling; load with pandas or datasets. License CC BY 4.0 (OpenADMET). Cite OpenADMET.","displayTitle":"OpenADMET pKa (SiriusT3): Experimental Aqueous pKa for 717 Small Molecules, UV-Metric and pH-Metric Titration","huggingFaceUrl":"https://huggingface.co/datasets/openadmet/pka-modeling-data","id":"9d467074-0ad6-5d3c-ac8e-c3d96e5aaaa4","license":"CC-BY-4.0","licenseEvidenceUrl":"https://openadmet.org/datasetsmodels/","objectCountEstimate":3,"probeStatus":"ok","region":"us","schemaCard":{"accessHints":[{"language":"python-datasets","snippet":"from datasets import load_dataset\nds = load_dataset(\"openadmet/pka-modeling-data\")"},{"language":"python-hub","snippet":"from huggingface_hub import snapshot_download\nsnapshot_download(repo_id=\"openadmet/pka-modeling-data\", repo_type=\"dataset\")"}],"caveats":["HuggingFace metadata: author=openadmet, license:cc-by-4.0, size_categories:1K<n<10K, format:csv, downloads=292"],"format":"mixed","generatedAt":"2026-11-22T01:10:22.535Z","layout":{"example":"pka_measurements.csv","pattern":"{repoId}/{filename}"},"mountPath":"/mnt/archive","sampleHeaders":[{"bytes":"manual_QC,pKa_model,Enamine_ID,SMILES,assay_method,pka_index,pka_value,pka_std,pka_ionic_strength,pka_temperature,cosolvent,cosolvent_fractions,comment_code\n0,1,EN300-114893,O=CC1=CC(C2=NC=CC=C2)=CC=C1,UV-metric,1,4.07725,0.0530671,0.193084,24.9731,Methanol,\"[0.464403, 0.342225, 0.247079]\",0\n0,1,EN300-17336,CC1=C(O)C(C=O)=C(CO)C=N1,UV-metric,1,3.23395,0.224595,0.192631,25.0519,Methanol,\"[0.465991, 0.346165, 0.251549]\",0\n0,1,EN300-725703,CN(C)C1=C(C=O)C=C([N+]([O-])=O)C=C1,UV-metric,1,0.861051,0.0363434,0.19","encoding":"utf8","key":"pka_measurements.csv","truncatedAt":512}],"sampleObjects":["pka_measurements.csv"],"sidecars":[{"bytes":"---\nlicense: cc-by-4.0\nlanguage:\n- en\ntags:\n- chemistry\n- drug-discovery\n- physicochemical-properties\n- pKa\n- protonation\n- molecular-properties\npretty_name: OpenADMET pKa (SiriusT3)\nsize_categories:\n- 1K<n<10K\ntask_categories:\n- tabular-regression\nannotations_creators:\n- expert-generated\nsource_datasets:\n- original\nconfigs:\n- config_name: default\n  data_files:\n  - split: train\n    path: pka_measurements.csv\n---\n\n# OpenADMET pKa (SiriusT3)\n\nExperimentally measured aqueous **pKa** values for 717 small molecules, produced through the\nOpenADMET-Schrodinger collaboration.\n\npKa describes the pH-dependent equilibrium between protonation states differing by a single proton,\nand is one of the most fundamental descriptors of a small molecule — it shapes expectations for\nsolubility, permeability and broader ADME behaviour. The compound set was chosen deliberately to\ninclude chemistry where prediction is hardest: tautomerisable compounds with uncertain equilibria,\nelectron-deficient heteroaromatics, and heterocycles with sparse literature precedent, alongside\nmore familiar carboxylic acids, phenols and substituted pyridines. Much of it is measured here for\nthe first time.\n\nA reported pKa is **not a direct instrument readout**. It is the endpoint of a chain of\nmeasurement, modelling and interpretation: titration, extrapolation to zero cosolvent, and the\nchoice of how many protonation transitions to fit. This release exposes that last step rather than\nhiding it — see [Choosing a fit](#choosing-a-fit-read-this-first).\n\n- **Blog post:** _link to be added once published_\n- **Produced by:** OpenADMET and Schrodinger\n- **Instrument:** Pion SiriusT3\n\n## Dataset at a glance\n\n| | |\n|---|---|\n| Rows | 7,246 |\n| Compounds | 717 (unique `Enamine_ID`, one-to-one with `SMILES`) |\n| Rows per `pKa_model` | 1 pKa: 744 rows · 2 pKa: 1,466 rows · 3 pKa: 2,160 rows · 4 pKa: 2,876 rows |\n| Assay method | UV-metric 7,222 rows; pH-metric 24 rows |\n| Fits within the pH 2-12 titration window | 79.5% |\n| Rows with manual review (`manual_QC == 1`) | 56 |\n| Rows carrying a `comment_code` | 636 |\n\n## Choosing a fit (read this first)\n\nEach measurement was fitted with **1-, 2-, 3- and 4-pKa models, and every fit is reported**. No\npreferred interpretation has been selected. A compound measured once therefore contributes\n1 + 2 + 3 + 4 = 10 rows, and `pka_value` is only meaningful once you decide how many protonation\ntransitions the molecule actually undergoes in the titration range.\n\n**Do not train directly on the `pka_value` column.** Select a `pKa_model` per compound first.\n\nPicking that model requires an estimate of how many transitions are expected between pH 2 and 12.\nIn the source work this came from Schrodinger's Macro-pKa workflow, which also reports tautomeric\nmicrostate populations; the predicted transition count then determined which fitting model to\ntrust. Where several models look mathematically plausible, that external prediction is often the\nonly way to choose correctly.\n\nSignals in the table that help narrow it down:\n\n- `pka_std` — the fitted uncertainty. It climbs sharply with `pKa_model` as extra parameters go\n  unconstrained (median 0.053 at 1 pKa versus\n  0.797 at 4).\n- Whether `pka_value` falls inside pH 2-12. Outside that range the value is extrapolated, not\n  observed. 79.5% of all rows fall inside; the fraction drops as `pKa_model` rises.\n- `comment_code` and `manual_QC` — flags from the experimental review.\n\nAs a rough indication of coverage, 710 of the 717\ncompounds have at least one fit that is both inside the window and has `pka_std < 1`.\n\n### Repeated keys\n\nRows are **not** unique on `(Enamine_ID, pKa_model, pka_index)`. Where a key repeats, the rows are\nalternative processings of the same transition — typically an original fit alongside a manually\nreviewed one (`manual_QC == 1`, often with `comment_code` 5), or a UV-metric screen alongside a\npH-metric confirmation. Compare `manual_QC`, `assay_method` and `comment_code` to pick between\nthem; the reviewed fit is generally the better one.\n\n## How to get the data\n\n### With `da","encoding":"utf8","key":"README.md","name":"README.md","size":10220,"truncated":true}],"sourceObjectKey":"pka_measurements.csv","starterCell":"import pandas as pd\ndf = pd.read_csv(\"hf://datasets/openadmet/pka-modeling-data/pka_measurements.csv\")\ndf.head()","starterSamplePath":"pka_measurements.csv","totalObjects":3},"summary":"Experimentally measured aqueous acid dissociation constants (pKa) for 717 small molecules from the OpenADMET and Schrodinger collaboration, measured on a Pion SiriusT3.","title":"OpenADMET pKa (SiriusT3): Experimental Aqueous pKa for 717 Small Molecules, UV-Metric and pH-Metric Titration","updatedAt":"2026-11-22T04:02:26.772Z","visibility":"public","websiteUrl":"https://huggingface.co/datasets/openadmet/pka-modeling-data"}}