OpenADMET ExpansionRx Full ADMET Dataset: sample notebook
A worked notebook for OpenADMET ExpansionRx Full ADMET Dataset, with 4 code cells and the output each one printed. On Scigantic it opens in a hosted JupyterLab with the dataset already mounted, so nothing is downloaded first.
- Provider
- OpenADMET
- License
- cc-by-4.0
- Format
- CSV (in sampled files)
- Stored at
- Hugging Face (https://huggingface.co/datasets/openadmet/openadmet-expansionrx-challenge-data)
- Access
- Free and public. No account with the data provider is needed.
The notebook
OpenADMET ExpansionRx Full ADMET Dataset — Getting Started
Real ADMET measurements from a live Expansion Therapeutics drug-discovery program on RNA-mediated diseases — the finalized, post-challenge release (the blind challenge closed January 19, 2026 with 370+ participants). Nine endpoints across solubility, lipophilicity, clearance, permeability, and tissue binding.
The one thing to see before modeling anything here: this table is sparse, and not randomly so.
from datasets import load_dataset
import pandas as pd
df = load_dataset("openadmet/openadmet-expansionrx-challenge-data", "default", split="train").to_pandas()
print(f"{len(df)} compounds")
df.head()
Output:
5326 compounds Molecule Name SMILES LogD \ 0 E-0001321 CN1CCC[C@H]1COc1ccc(-c2nc3cc(-c4ccc5[nH]c(-c6c... NaN 1 E-0001780 COc1ccc2c(c1)c1cc3cnccc3c(C)c1n2C NaN 2 E-0001827 Cc1c2ccncc2cc2c3cc(OCCCN4CCN(C)CC4)ccc3n(C)c12 NaN 3 E-0002019 CN(C)CCCOc1ccc(-c2nc3cc(NC(=O)c4ccc5[nH]c(-c6c... NaN 4 E-0002036 CN(C)CCCOc1ccc2nc(-c3ccc(-c4nc5ccc(OCCCN(C)C)c... NaN KSOL HLM CLint MLM CLint Caco-2 Permeability Papp A>B \ 0 NaN 56.4 182.3 NaN 1 NaN 160.4 1351.1 NaN 2 NaN NaN 193.5 NaN 3 NaN 9.5 NaN NaN 4 NaN NaN 162.0 NaN ...
Endpoint coverage: "whack-a-mole" measurement, not a dense matrix
Each compound was measured only for the endpoints that mattered at that moment in the campaign, so coverage swings from ~96% down to under 6%. dropna() across all nine endpoints at once leaves almost nothing — train per-endpoint instead.
endpoints = ["LogD", "KSOL", "HLM CLint", "MLM CLint", "Caco-2 Permeability Papp A>B",
"Caco-2 Permeability Efflux", "MPPB", "MBPB", "MGMB"]
import matplotlib.pyplot as plt
coverage = df[endpoints].notna().sum().sort_values()
fig, ax = plt.subplots(figsize=(8, 4.5))
coverage.plot.barh(color="steelblue", ax=ax)
ax.set_xlabel(f"Compounds with a measurement (of {len(df)})")
ax.set_title("Per-endpoint coverage")
plt.tight_layout()
One endpoint, close up
HLM CLint (human liver microsomal intrinsic clearance, mL/min/kg) is the densest clearance endpoint — higher means the liver breaks the compound down faster.
df["HLM CLint"].dropna().hist(bins=40, figsize=(7, 4), color="steelblue")
plt.xlabel("HLM CLint (mL/min/kg)"); plt.title(f"n={df['HLM CLint'].notna().sum()}")
plt.tight_layout()
What the compounds look like
from rdkit import Chem
from rdkit.Chem import Draw
def draw(smiles_list, legends, per_row=4):
mols, labels = [], []
for s, l in zip(smiles_list, legends):
m = Chem.MolFromSmiles(s)
if m is not None:
mols.append(m)
labels.append(l)
return Draw.MolsToGridImage(mols, molsPerRow=per_row, subImgSize=(220, 180), legends=labels)
sample = df.dropna(subset=["HLM CLint"]).sample(8, random_state=0)
draw(sample["SMILES"], [f"CLint={v:.0f}" for v in sample["HLM CLint"]])
Where to go next
- raw config keeps out-of-range censored measurements (" 100") that default drops — use it for censored-aware modeling instead of pd.to_numeric(errors="coerce"), which silently discards them.
- MPPB/MBPB/MGMB are % unbound (high = less bound) — the opposite direction from a typical "percent bound" column elsewhere.
- Don't join this to PubChem on SMILES — these are optimized medicinal-chemistry structures in a different canonical convention; under 1% match exactly.
- Open this archive's schema card for unit-by-unit detail on every endpoint.
Run it
Open this notebook to run it, see the dataset page for the full description, or browse the full catalog.