pytdc

What it is and what it does

PyTDC is a machine-learning platform built on the Therapeutic Data Commons that unifies heterogeneous biomedical datasets, model weights, and benchmarking infrastructure for drug discovery research. It implements an API-first architecture to provide standardized access to multimodal biological data—including molecular, protein, and single-cell information—alongside curated machine learning tasks spanning target discovery, activity screening, efficacy, and safety evaluation across small molecules, antibodies, and vaccines.

The package streamlines the full ML workflow: data loading and preprocessing, model training and evaluation against established benchmarks, and inference using state-of-the-art research-ready models. It integrates single-cell analysis with multimodal machine learning, enabling contextualized tasks that combine foundation models with therapeutic applications. Most datasets can be retrieved with minimal code, and the platform provides data functions for splitting, evaluation, and molecule generation—reducing boilerplate and standardizing how researchers compare methods.

Use it for:

• Benchmark ML models on standardized drug discovery tasks to compare against published baselines.
• Access curated, preprocessed molecular and biological datasets without writing custom data pipelines.
• Train and evaluate single-cell foundation models for transfer learning in therapeutic applications.
• Retrieve and deploy pre-trained biomedical representation learning models for inference.
• Develop and validate new ML methods for drug safety, efficacy, and manufacturing prediction.

Worth the install?

AI-flagged interpretation of the facts on this page — verify before relying

PyTDC provides unified access to multimodal biomedical datasets, machine learning benchmarks, and state-of-the-art model weights for drug discovery and therapeutic AI research, with standardized training, evaluation, and inference endpoints.

Yes, with conditions. PyTDC is worth installing if you are conducting drug discovery or therapeutic AI research and need standardized, multimodal datasets and benchmarks. The high install friction (19 dependencies, including rdkit and specialized bioinformatics libraries) and 501-day gap since last release are trade-offs; verify that your Python environment and system can support the full dependency stack, and check the repository for current maintenance status before committing to production workflows. No known security vulnerabilities and permissive MIT licensing are strong positives.

Install

pytdc on PyPI

pip install pytdc

uv add pytdc

poetry add pytdc

Installing pytdc

Before you install

High install friction: 19 runtime dependencies including transformers, datasets, huggingface_hub, scikit-learn, rdkit, and specialized bioinformatics libraries (cellxgene-census, tiledbsoma, gget). Package is aging (501 days since last release) but repository remains active with recent commits and 1272 stars.

License in practice

MIT license permits commercial and private use with minimal restrictions, making it suitable for both research and production deployment in therapeutic applications.

Quickstart

pip install pytdc
from pytdc import DataLoader
data = DataLoader(name='DATASET_NAME')
X, y = data.get_split(split='train')

Requires rdkit (chemistry toolkit) and cellxgene-census/tiledbsoma (single-cell data access), which may need system-level dependencies or network access to remote data repositories.

Verify before relying

• Minimum Python version requirement not specified in metadata; compatibility with modern Python versions unclear.
• Whether all 19 runtime dependencies are required for basic usage or if subsets can be installed for specific tasks.
• Current maintenance cadence and responsiveness to issues given 501-day gap since last release.

Package facts

Evidence: pytdc-1.1.15.tar.gz

Tags

More Artificial Intelligence packages