How do I evaluate my models with PyTDC?

PyTDC provides an Evaluator class with standardized metrics like ROC-AUC, RMSE, and MAE that are specifically mapped to each task's requirements.

PyTDC is the Therapeutics Data Commons, an open-science library providing AI-ready datasets and benchmarks for the entire drug discovery and development pipeline.

Does PyTDC include toxicity datasets?

Yes, it includes several curated toxicity datasets such as hERG (cardiotoxicity), AMES (mutagenicity), and DILI (drug-induced liver injury).

Can I use PyTDC for molecule generation?

Yes, PyTDC supports molecular generation tasks and provides 'oracles' to evaluate and optimize generated molecules for specific chemical properties.

What kind of data splits does PyTDC support?

It supports random, scaffold (based on chemical diversity), and 'cold' splits (unseen drugs or targets) to ensure models generalize to novel chemical spaces.

PyTDC Therapeutics Data Commons

Name: PyTDC Therapeutics Data Commons
Author: Zehong-Wang

byZehong-Wang

0•

データサイエンスとML

Facilitates drug discovery and therapeutic machine learning by providing AI-ready datasets, benchmarks, and molecular oracles.

PyTDC (Therapeutics Data Commons) is a comprehensive open-science platform designed to streamline the application of machine learning in the therapeutics pipeline. It offers a curated collection of AI-ready datasets spanning absorption, distribution, metabolism, excretion, and toxicity (ADMET), along with tools for drug-target interaction (DTI) prediction and novel molecule generation. By providing standardized evaluation metrics, diverse data splitting strategies—such as scaffold and cold-start splits—and molecular oracles for property optimization, PyTDC enables researchers to build, benchmark, and deploy robust models for pharmacological prediction and drug development tasks directly within their workflow.

主な機能

01Access to curated ADMET, Toxicity, and High-Throughput Screening datasets

02Advanced data splitting methods including scaffold and cold-target splits

03Standardized benchmarking with multi-seed evaluation protocols

04Integration with 17+ molecular oracles for property-driven generation

050 GitHub stars

06Built-in utilities for molecular format conversion and entity retrieval

ユースケース

01Predicting drug-target interactions and binding affinities for lead discovery

02Optimizing molecular structures for desired pharmacokinetic properties using oracle functions

03Benchmarking new machine learning architectures on standardized pharmaceutical datasets

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add zehong-wang/kosmos pytdc

For use in Claude.ai and ChatGPT

Download Skill