Datasets:

openadmet
/

cyp-challenge-train-test

license: apache-2.0
language:
  - en
tags:
  - chemistry
  - drug-discovery
  - ADMET
  - molecular-properties
  - blind-challenge
  - computational-chemistry
  - cytochrome-p450
pretty_name: OpenADMET CYP Inhibition Blind Challenge
size_categories:
  - 10K<n<100K
task_categories:
  - tabular-regression
annotations_creators:
  - expert-generated
source_datasets:
  - original
configs:
  - config_name: default
    data_files:
      - split: train
        path: cyp-challenge-TRAIN_inhibition.csv
      - split: test
        path: cyp-challenge-TEST-BLINDED.csv
  - config_name: tdi
    data_files:
      - split: train
        path: cyp-challenge-TRAIN_TDI.csv
  - config_name: single_concentration
    data_files:
      - split: train
        path: cyp-challenge-single-concentration-TRAIN.csv
  - config_name: emax
    data_files:
      - split: train
        path: cyp-challenge-TRAIN_Emax.csv

CYP Challenge Train/Test Dataset

A high-quality experimental dataset for predicting inhibition of the major drug-metabolizing Cytochrome P450 enzymes (CYP1A2, CYP2C9, CYP2D6, CYP3A4), released as part of the OpenADMET CYP Inhibition Blind Challenge.

Blog post: Announcing OpenADMET’s CYP inhibition blind challenge

Challenge Space: OpenADMET CYP Inhibition Blind Challenge

Challenge period: August 17, 2026 - November 3, 2026

Produced by: OpenADMET

CHANGELOG

Updated 2026-08-06 Finalised ground truth: refreshed direct-inhibition and TDI training/test data with the finalised assay results, and added a new Emax (maximal effect) training set.
Updated 2026-08-03 Initial release: direct inhibition training/blinded test data, time-dependent inhibition (TDI) training data, and single-concentration screening training data.

Dataset contents

Config	Split	File	Description
`default`	`train`	`cyp-challenge-TRAIN_inhibition.csv`	Primary direct-inhibition training set — 4,905 compounds with pIC50 (plus 95% CI and std) for CYP1A2, CYP2C9, CYP2D6, CYP3A4
`default`	`test`	`cyp-challenge-TEST-BLINDED.csv`	750-compound blinded test set (SMILES only; labels withheld for the challenge)
`tdi`	`train`	`cyp-challenge-TRAIN_TDI.csv`	Time-dependent inhibition (TDI) training set — 6,145 compounds with `is_TDI` classification labels for CYP2D6/CYP3A4, TDI-condition pIC50s, and paired direct-inhibition pIC50s for comparison
`single_concentration`	`train`	`cyp-challenge-single-concentration-TRAIN.csv`	Single-concentration screening data — 17,504 measurements across 4,376 compounds x 4 enzymes (log2 fold-change format)
`emax`	`train`	`cyp-challenge-TRAIN_Emax.csv`	Emax (maximal effect) training set — 6,146 compounds with `is_TDI` classification labels for all four CYPs, plus TDI-condition and direct-inhibition Emax values (with 95% CI)

Loading with Hugging Face `datasets`

from datasets import load_dataset

# Default config (primary direct-inhibition assay)
ds = load_dataset("openadmet/cyp-challenge-train-test")
train = ds["train"]
test  = ds["test"]

# TDI (time-dependent inhibition) config
ds_tdi = load_dataset("openadmet/cyp-challenge-train-test", "tdi")
train_tdi = ds_tdi["train"]

# Single-concentration config
ds_single = load_dataset("openadmet/cyp-challenge-train-test", "single_concentration")
train_single = ds_single["train"]

# Emax config
ds_emax = load_dataset("openadmet/cyp-challenge-train-test", "emax")
train_emax = ds_emax["train"]

Loading directly with pandas

import pandas as pd

train        = pd.read_csv("hf://datasets/openadmet/cyp-challenge-train-test/cyp-challenge-TRAIN_inhibition.csv")
test         = pd.read_csv("hf://datasets/openadmet/cyp-challenge-train-test/cyp-challenge-TEST-BLINDED.csv")
train_tdi    = pd.read_csv("hf://datasets/openadmet/cyp-challenge-train-test/cyp-challenge-TRAIN_TDI.csv")
train_single = pd.read_csv("hf://datasets/openadmet/cyp-challenge-train-test/cyp-challenge-single-concentration-TRAIN.csv")
train_emax   = pd.read_csv("hf://datasets/openadmet/cyp-challenge-train-test/cyp-challenge-TRAIN_Emax.csv")

CYP Challenge Train/Test Dataset

CHANGELOG

Dataset contents

Loading with Hugging Face datasets

Loading directly with pandas

Loading with Hugging Face `datasets`