Therapeutics Data Commons (TDC)

1.1.15 verified Fri May 01 auth: no python

Therapeutics Data Commons (TDC) is a unified, open-source framework for machine learning in therapeutics. It provides standardized datasets, benchmarks, and tools for tasks like drug-target interaction, ADMET prediction, and clinical trial outcome prediction. Current version 1.1.15, with frequent updates.

pip install PyTDC

Common errors

error ModuleNotFoundError: No module named 'tdc' ↓

cause pyTDC not installed or installed in wrong environment.

fix

Run: pip install PyTDC

error AttributeError: module 'tdc' has no attribute 'single_pred' ↓

cause Attempting to import from top-level tdc incorrectly.

fix

Use: from tdc.single_pred import ADME

error urllib.error.URLError: <urlopen error [Errno 111] Connection refused> ↓

cause Network issue when downloading datasets.

fix

Check internet connection or set TDC_CACHE_DIR to a local path. Alternatively, use a local data file with DataLoader.

Warnings

breaking Python 3.7 support dropped after version 0.9.6. Use Python 3.8+. ↓

fix Upgrade Python to 3.8 or higher.

gotcha Data download may fail behind corporate firewall or with slow network. TDC downloads from remote servers. ↓

fix Set environment variable TDC_CACHE_DIR to a writable directory, or use data.DataLoader for custom files.

gotcha The split method 'random' produces same splits across runs without seed? Actually you must pass seed explicitly for reproducibility. ↓

fix Always provide seed parameter to get_split(method='random', seed=42).

Imports

TDC
```
from tdc import TDC
```
Main entry point for accessing datasets.
SinglePredDataset
wrong
```
from tdc import SinglePredDataset
```
correct
```
from tdc.single_pred import SinglePredDataset
```
Older imports from top-level tdc may not work; use submodule.
DataLoader
```
from tdc.utils import DataLoader
```
Utility for loading local files.

Quickstart

Load Caco2 permeability dataset and split into train/valid/test.

from tdc.single_pred import ADME
data = ADME(name = 'Caco2_Wang')
split = data.get_split(method = 'random', seed = 42)
train, valid, test = split['train'], split['valid'], split['test']
print(train.head())