What datasets are available in the PyTDC skill?

PyTDC provides access to a wide range of therapeutic datasets including ADME properties, toxicity (Tox), Drug-Target Interactions (DTI), and molecular generation tasks like Retrosynthesis.

How does this skill help with model evaluation?

It provides standardized Benchmark Groups with pre-defined evaluation protocols and metrics like ROC-AUC or RMSE, ensuring results are comparable across different research papers.

Does it support chemical structure processing?

Yes, it includes utilities for converting between SMILES, SELFIES, and graph representations (PyG/DGL), as well as filters for drug-likeness and PAINS.

What is a 'scaffold split' in PyTDC?

A scaffold split is a data partitioning method that ensures structurally diverse molecules are placed in different sets (train/test), providing a more realistic assessment of a model's ability to generalize to new chemical space.

PyTDC Therapeutics Data Commons

Name: PyTDC Therapeutics Data Commons
Author: pathvanasoftware

bypathvanasoftware

데이터 과학 및 ML

Provides AI-ready datasets and benchmarks for drug discovery, including ADME, toxicity, and molecular generation tasks.

소개

The PyTDC skill equips Claude with specialized knowledge for the Therapeutics Data Commons, an open-science platform for drug discovery machine learning. It streamlines the retrieval of curated datasets for pharmacokinetic properties, binding affinities, and drug interactions, while providing standardized evaluation metrics and scaffold-based data splits. Whether you are benchmarking new models on ADMET properties or optimizing novel molecules using molecular oracles, this skill automates the boilerplate code for data loading and processing in pharmaceutical ML workflows.

주요 기능

0 GitHub stars
Access to curated ADME, toxicity, and bioactivity datasets
Standardized benchmark groups for systematic model evaluation
Implementation of scaffold, cold-split, and temporal data splitting
Molecular oracles for property optimization and generation
Data utilities for SMILES conversion and chemical property filtering

사용 사례

Benchmarking machine learning models on standardized pharmaceutical datasets
Predicting molecular properties like solubility, permeability, and toxicity
Optimizing lead compounds through goal-directed molecular generation

소개

주요 기능

0 GitHub stars
Access to curated ADME, toxicity, and bioactivity datasets
Standardized benchmark groups for systematic model evaluation
Implementation of scaffold, cold-split, and temporal data splitting
Molecular oracles for property optimization and generation
Data utilities for SMILES conversion and chemical property filtering

사용 사례

Benchmarking machine learning models on standardized pharmaceutical datasets
Predicting molecular properties like solubility, permeability, and toxicity
Optimizing lead compounds through goal-directed molecular generation