What kind of data splits does PyTDC support?

It supports specialized splits such as scaffold splits for chemical diversity and cold-drug/cold-target splits to test model generalization on previously unseen entities.

What datasets are included in PyTDC?

PyTDC includes curated datasets for ADME, toxicity, drug-target interactions (DTI), drug-drug interactions (DDI), molecule generation, and retrosynthesis.

How does this skill help with molecule generation?

It provides access to molecular oracles, which are scoring functions that evaluate molecules for specific properties like bioactivity or drug-likeness during the optimization process.

Can I use this for model benchmarking?

Yes, it includes specialized benchmark groups like the ADMET group, providing a standardized evaluation protocol with 5-seed testing and predefined splits.

PyTDC Therapeutics Data Commons

Name: PyTDC Therapeutics Data Commons
Author: ricable

byricable

•

データサイエンスとML

Accesses and manages AI-ready drug discovery datasets, benchmarks, and molecular oracles for therapeutic machine learning.

PyTDC is an open-science platform designed to streamline drug discovery and development by providing standardized, AI-ready datasets across the entire therapeutics pipeline. It offers specialized tools for predicting molecular properties like ADME and toxicity, modeling drug-target interactions, and generating novel molecules with specific desired traits. By integrating benchmarks and specialized data split strategies like scaffold and cold-start splits, it ensures that machine learning models are evaluated rigorously for real-world pharmaceutical applications.

主な機能

01Standardized evaluation metrics for pharmacokinetics and toxicity prediction

02Access to 100+ AI-ready drug discovery and therapeutic datasets

03Integration with 17+ molecular oracles for goal-directed generation

04Automated data processing and format conversion for PyG, DGL, and SMILES

051 GitHub stars

06Advanced data splitting strategies including scaffold and cold-target splits

ユースケース

01Benchmarking drug-target interaction (DTI) models on standardized datasets

02Predicting ADME and toxicity profiles for new drug candidates

03Optimizing molecular generation for specific biological targets using oracles

主な機能

01Standardized evaluation metrics for pharmacokinetics and toxicity prediction

02Access to 100+ AI-ready drug discovery and therapeutic datasets

03Integration with 17+ molecular oracles for goal-directed generation

04Automated data processing and format conversion for PyG, DGL, and SMILES

051 GitHub stars

06Advanced data splitting strategies including scaffold and cold-target splits

ユースケース

01Benchmarking drug-target interaction (DTI) models on standardized datasets

02Predicting ADME and toxicity profiles for new drug candidates

03Optimizing molecular generation for specific biological targets using oracles