How does PyTDC handle data splitting for chemistry?

It supports advanced, domain-relevant splitting strategies such as scaffold-based splits for chemical diversity and 'cold-split' methods for evaluating performance on unseen drugs or targets.

What kind of datasets does the PyTDC skill provide?

It provides curated, AI-ready datasets for single-instance prediction (ADME, toxicity, HTS, QM), multi-instance prediction (DTI, DDI, PPI), and molecular generation tasks.

Can I use this skill for molecular property optimization?

Yes, the skill includes access to over 17 molecular oracle functions that can be used to score and optimize molecules for specific biological and chemical properties.

What are PyTDC Benchmark Groups?

Benchmark Groups are curated collections of datasets (like the ADMET group) with standardized evaluation protocols and metrics designed to enable fair model comparison.

PyTDC (Therapeutics Data Commons)

Name: PyTDC (Therapeutics Data Commons)
Author: x-cmd

byx-cmd

•

Ciencia de Datos y ML

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

PyTDC is an open-science platform designed to streamline drug discovery workflows by providing access to high-quality, curated datasets and standardized evaluation metrics. It covers the entire therapeutics pipeline, including molecular property prediction, drug-target interactions, and molecule generation. By offering specialized data splits like scaffold and cold-start, along with built-in molecular oracles for property optimization, PyTDC enables researchers to benchmark machine learning models accurately and accelerate the development of novel pharmacological treatments.

Características Principales

018 GitHub stars

02Apply domain-specific data splits including scaffold, temporal, and cold-start strategies

03Implement standardized benchmark groups with 5-seed evaluation protocols

04Predict drug-target (DTI) and drug-drug (DDI) interactions with multi-instance learning

05Access curated datasets for ADME, toxicity, and bioactivity prediction

06Utilize molecular oracles for goal-directed molecule generation and optimization

Casos de Uso

01Predicting pharmacokinetic properties like intestinal permeability and blood-brain barrier penetration

02Benchmarking machine learning architectures on standardized pharmaceutical datasets

03Generating and optimizing novel molecules with specific binding affinities and safety profiles

Características Principales

018 GitHub stars

02Apply domain-specific data splits including scaffold, temporal, and cold-start strategies

03Implement standardized benchmark groups with 5-seed evaluation protocols

04Predict drug-target (DTI) and drug-drug (DDI) interactions with multi-instance learning

05Access curated datasets for ADME, toxicity, and bioactivity prediction

06Utilize molecular oracles for goal-directed molecule generation and optimization

Casos de Uso

01Predicting pharmacokinetic properties like intestinal permeability and blood-brain barrier penetration

02Benchmarking machine learning architectures on standardized pharmaceutical datasets

03Generating and optimizing novel molecules with specific binding affinities and safety profiles