Pharmacokinetics Dataset · gxl-ai
All-Route Terminal Half-Life Dataset
Terminal pharmacokinetic half-lives, in hours, curated from 3,270 PubMed Central articles.
What All-Route Terminal Half-Life Dataset is
The All-Route Terminal Half-Life Dataset is a curated collection of terminal pharmacokinetic half-life observations reported in the biomedical literature. It holds 19,952 records drawn from 3,270 PubMed Central articles and spans oral, intravenous, subcutaneous and other administration routes. GXL (Generative Expert Labs) built it with Paperclip Datasets, its tool for turning scientific literature into structured datasets.
Each row is one observation, specific to a paper, analyte, regimen, cohort, sampling occasion and summary statistic. The half-life is normalised to hours so records can be compared, while the value as the paper reported it, with its uncertainty or dispersion, is kept as text beside it. Where a structure could be assigned, a row also carries the analyte's SMILES and InChIKey: 12,434 records have both.
The analytes are broader than parent small-molecule drugs. They include metabolites, antibodies, proteins, peptides, enzymes, oligonucleotides and total-radioactivity measurements, and every record carries one coarse molecule class saying which it is.
How it was made
Candidate terminal half-life observations were extracted from biomedical articles identified by PMCID, then reviewed for terminal-phase qualification, analyte identity, route, regimen, cohort, sampling occasion, summary statistic and units. Source-reported ranges were not averaged and no point estimates were invented; values were converted to hours, and the source-facing text, dispersion and intervals were retained.
Duplicates were adjudicated on analyte, dose, regimen, cohort, PK occasion, statistic and source-value context, not on equal numbers alone. LLM-assisted comparison of 218 candidate pairs across 213 high-confidence groups removed 215 redundant rows. Chemical identifiers were checked as paired SMILES and InChIKey values, and automated validation covered row identifiers, PMCIDs, required fields, half-life positivity, identifier formatting and controlled vocabularies.
Key facts
- Records
- 19,952
- Source articles
- 3,270 PubMed Central articles
- Columns
- 14
- Routes
- Oral 10,002; intravenous 7,705; subcutaneous 820; other 1,425
- Structures
- 12,434 records with paired SMILES and InChIKey
- Normalised drug names
- 16,874 records
- Molecule classes
- Small molecule 14,904; antibody 2,564; protein 854; peptide 620; other 401; antibody-drug conjugate 175; enzyme 172; oligonucleotide 157; oligosaccharide 105
- Half-life unit
- Hours
- Format
- CSV, one file of about 34 MB
- Release
- 2026-10-26.2
What you can use it for
- Pharmacokinetic literature review
- Find reported terminal half-lives by analyte and route, with the PMCID of the source article on every row so each value can be checked against its paper.
- Comparing analytes
- Compare half-lives across compounds once stratified by route, molecule class and summary statistic, for example antibodies against small molecules.
- Exploratory analysis
- Look at how reported terminal half-lives are distributed across routes and molecule classes, and generate hypotheses to test against the primary literature.
- Benchmarking information extraction
- Use the curated records as a reference for systems that extract, normalise and deduplicate pharmacokinetic values from papers.
Explore the data
No licence declared
The authors have not published a licence, which means no permission to redistribute the data has been granted to anyone. The rows are not served here; read them on the original dataset page.Strengths and limits
Good at
- Every record links to its source article by PMCID.
- The value as reported, its dispersion and the study context are kept as text beside the numeric hours, so nothing is flattened away.
- Observations are specific to regimen, cohort, sampling occasion and statistic, and equal numbers alone were not treated as duplicates.
- Structures are paired SMILES and InChIKey values, so records can be grouped by chemistry where an identity was resolved.
Watch out for
- Methods, sampling duration, matrices, terminal-phase definitions and model assumptions vary across papers, so values from different records are not automatically comparable.
- The analytes are not only parent drugs: metabolites, biomarkers, total radioactivity, antibodies, proteins and transgenes are included, so drug_name should not be read as a parent small molecule.
- 3,078 records lack a normalised drug name and 7,518 lack SMILES and InChIKey, since structures are often unavailable or inappropriate for mixtures, biologics, cells, polymers and radioactivity measurements.
- 317 records have an unspecified summary statistic, and some values reflect absorption-, depot-, implant-, transgene-, cellular-kinetic- or formulation-limited behaviour rather than elimination.
- It is not clinical guidance and should not be used alone for prescribing, dose selection, therapeutic monitoring or regulatory decisions.
Questions
Can I look inside All-Route Terminal Half-Life Dataset without downloading it?
Not here: it declares no licence, so no permission to redistribute its rows has been granted. Its description and statistics are still shown.
What licence is All-Route Terminal Half-Life Dataset under?
Its authors have not declared a licence, which means no permission to reuse or redistribute it has been granted.
What does one row represent?
One terminal half-life observation, specific to a paper, analyte, regimen, cohort, sampling occasion and summary statistic. terminal_half_life gives the value in hours, and reported_half_life keeps it as the paper stated it, with units and any dispersion or interval.
Which columns does it have?
Fourteen: row_id, pmcid, drug_name, normalized_drug_name, molecule_class, smiles, inchikey, administration_route, terminal_half_life, reported_half_life, summary_statistic, drug_information, cohort_information and notes. Blank strings mean a value is unavailable, not zero.
Are all the analytes small-molecule drugs?
No. 14,904 records are small molecules; the rest are antibodies (2,564), proteins (854), peptides (620), other analytes (401), antibody-drug conjugates (175), enzymes (172), oligonucleotides (157) and oligosaccharides (105). molecule_class describes the measured analyte, so the same drug can appear under different classes, for example free and polymer-conjugated.
How should I treat records with an unspecified statistic?
317 records have summary_statistic set to unspecified, meaning the paper does not say which statistic governs the value. Read reported_half_life and notes before treating them as conventional point estimates. The release contains no range-endpoint records.
Does it come with training and test splits?
No. The Hugging Face table is exposed as a single train split so the datasets library can load it, but that is a storage convention: the dataset has no official validation or test partitions.
How to cite it
GXL. All-Route Terminal Half-Life Dataset. Version 2026-10-26.2. Hugging Face Datasets. SHA-256 89584232b5fbb52bf08a14c3c8fecabc818646b33e5fc7600a8447778ae21a90.
Written from All-Route Terminal Half-Life Dataset's own dataset card. Statements citing a figure that does not appear there are removed before publication. Licence and running details come from our registry.