Drug Discovery · openadmet
openadmet-expansionrx-challenge-train-data
Training dataset containing real-world ADMET endpoints from Expansion Therapeutics drug discovery campaigns.
What openadmet-expansionrx-challenge-train-data is
The openadmet-expansionrx-challenge-train-data dataset provides experimental absorption, distribution, metabolism, and excretion (ADME) measurements collected during preclinical lead optimisation by Expansion Therapeutics. The data originated from drug discovery campaigns focused on RNA-mediated diseases, where researchers evaluated drug candidates across off-target assays and physical property screens.
Organised in collaboration with OpenADMET for a blind predictive challenge, the dataset addresses the common medicinal chemistry challenge of forecasting late-stage compound properties using early-stage screening outcomes. It provides computational researchers with an empirical, industry-derived benchmark to train and validate predictive ADMET models.
How it was made
The data reflects experimental outcomes across nine preclinical parameters: distribution coefficient (LogD), kinetic solubility (KSOL in µM), intrinsic clearance in mouse and human liver microsomes (MLM and HLM CLint in mL/min/kg), Caco-2 cell permeability (Papp A>B and efflux ratio), and the percentage of unbound compound across mouse plasma (MPPB), mouse brain (MBPB), and mouse gastrocnemius muscle (MGMB).
Measurements are presented in two distinct tabular formats. The standard machine-learning-ready file contains only in-range numerical values, whereas an alternative raw version includes experimental modifier symbols, such as greater-than signs, to retain out-of-range measurements as recorded during laboratory assays.
Key facts
- Data provider
- Expansion Therapeutics and OpenADMET
- Domain
- Drug discovery and medicinal chemistry
- Biological context
- RNA-mediated disease programmes
- Assay endpoints
- 9 ADMET endpoints
- Data format
- Tabular CSV
What you can use it for
- Lead optimisation property forecasting
- Train models on earlier-stage chemical screening outcomes to predict the ADMET characteristics of later-stage analogues.
- Metabolic clearance prediction
- Develop quantitative structure-activity relationship models for mouse and human liver microsomal clearance.
- Tissue distribution and protein binding modelling
- Predict the free fraction of drug candidates across plasma, brain, and skeletal muscle tissue.
- Censored data handling in cheminformatics
- Benchmark machine learning algorithms capable of incorporating bounded assay results using the raw dataset with modifier flags.
Explore the data
The schema, real rows, full-text search and a SQL filter — read in your browser, nothing downloaded.
Explore
Read the real rows without downloading anything
Reading rows…
LiveRead from openadmet/openadmet-expansionrx-challenge-train-data at the moment you asked. Nothing is cached or stored — every row above came from that request.
How the columns are distributed
Over 5,326 rows of default/train
Caco-2 Permeability Efflux
float- min
- 0.26
- median
- 1.38
- mean
- 4.22
- max
- 105.64
59.4% empty
Caco-2 Permeability Papp A>B
float- min
- 0
- median
- 9.82
- mean
- 12.40
- max
- 51.41
59.5% empty
HLM CLint
float- min
- 0
- median
- 16.40
- mean
- 52.78
- max
- 2589.90
29.4% empty
KSOL
float- min
- 2.9e-3
- median
- 131
- mean
- 146.34
- max
- 325
3.7% empty
LogD
float- min
- -2
- median
- 2.10
- mean
- 2.11
- max
- 5.20
5.4% empty
MBPB
float- min
- 0
- median
- 3.38
- mean
- 7.67
- max
- 99.23
81.7% empty
MGMB
float- min
- 0
- median
- 4.42
- mean
- 7.53
- max
- 61.80
95.8% empty
MLM CLint
float- min
- 0
- median
- 219.65
- mean
- 560.53
- max
- 10354.80
15.1% empty
MPPB
float- min
- 0
- median
- 8.90
- mean
- 14.75
- max
- 87.60
75.6% empty
Molecule Name
string_text- min
- 9
- median
- 9
- mean
- 9
- max
- 9
SMILES
string_text- min
- 22
- median
- 46
- mean
- 48.03
- max
- 88
Strengths and limits
Good at
- Contains empirical measurements from genuine industrial drug discovery campaigns rather than synthetic benchmark tasks.
- Provides both filtered in-range data and raw files retaining assay boundary modifiers.
- Captures nine diverse ADMET parameters covering solubility, lipophilicity, permeability, metabolism, and multiple-tissue binding.
Watch out for
- Data originates from specific chemical series targeting RNA-mediated diseases, which may limit structural diversity.
- The curated machine-learning-ready split excludes out-of-range censored values, omitting data points present in the raw assays.
- Contains exclusively training-phase data for the challenge, lacking accompanying validation holdouts in this release.
Questions
Can I look inside openadmet-expansionrx-challenge-train-data without downloading it?
Yes. This page reads its schema, real rows and per-column statistics in the browser, with full-text search and a SQL filter. Nothing is downloaded to your machine.
What licence is openadmet-expansionrx-challenge-train-data under?
Its authors declare cc-by-4.0. Read the terms on the original page before reusing the data.
Which ADMET endpoints are tracked in this dataset?
The dataset contains nine endpoints: LogD, kinetic solubility (KSOL), mouse liver microsomal clearance (MLM CLint), human liver microsomal clearance (HLM CLint), Caco-2 efflux ratio, Caco-2 Papp A>B, mouse plasma protein binding (MPPB), mouse brain protein binding (MBPB), and mouse gastrocnemius muscle binding (MGMB).
How do the standard and raw CSV versions differ?
The standard machine-learning-ready file retains only measurements that fell strictly within experimental assay ranges, while the raw file preserves out-of-range modifier flags such as inequality operators.
What is the primary objective of the associated OpenADMET challenge?
The challenge asks participants to predict the ADMET properties of late-stage candidate molecules using data generated during earlier stages of the same medicinal chemistry campaigns.
Written from openadmet-expansionrx-challenge-train-data'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.