💊

pytdc

Name: pytdc
Author: K-Dense-AI

آمن ⚙️ الأوامر الخارجية🌐 الوصول إلى الشبكة

Access Drug Discovery Datasets with PyTDC

متاح أيضًا من: davila7

Drug discovery researchers need standardized datasets for training ML models. PyTDC provides curated ADME, toxicity, and drug-target interaction datasets with proper train-test splits and evaluation oracles.

يدعم: Claude Codex Code(CC)

🥉 76 برونزي

تنزيل ZIP المهارة

رفع في Claude

اذهب إلى Settings → Capabilities → Skills → Upload skill

فعّل وابدأ الاستخدام

اختبرها

استخدام "pytdc". Load the AMES toxicity dataset and show me the data format

النتيجة المتوقعة:

Dataset loaded with 7,255 compounds for mutagenicity prediction
Columns include Drug_ID, Drug (SMILES), and Y (binary toxicity label)
Scaffold split applied: 5,078 train, 725 validation, 1,452 test molecules

استخدام "pytdc". Evaluate this molecule with the GSK3B oracle: CC(C)Cc1ccc(cc1)C(C)C(O)=O

النتيجة المتوقعة:

GSK3B binding score: 0.0234 (low predicted affinity)
This SMILES represents ibuprofen, not expected to inhibit GSK3B
Scores range from 0 to 1, with higher values indicating stronger predicted binding

التدقيق الأمني

آمن

v4 • 1/17/2026

This skill provides documentation and templates for PyTDC, a legitimate drug discovery dataset library. All 427 static findings are false positives caused by markdown code blocks containing Python examples (detected as shell backticks), scientific terminology (DRD2, GSK3B detected as C2 keywords), and molecular/cryptographic naming overlaps. No actual security risks present.

الملفات التي تم فحصها

3,184

الأسطر التي تم تحليلها

النتائج

إجمالي عمليات التدقيق

عوامل الخطر

⚙️ الأوامر الخارجية (339)

🌐 الوصول إلى الشبكة (15)

EVALUATION_OUTPUT.json:42 EVALUATION_OUTPUT.json:42 EVALUATION_OUTPUT.json:42 references/datasets.md:246 references/oracles.md:374 references/oracles.md:398 references/utilities.md:681 references/utilities.md:682 references/utilities.md:683 references/utilities.md:684 scripts/benchmark_evaluation.py:268 skill-report.json:6 SKILL.md:454 SKILL.md:455 SKILL.md:456

تم تدقيقه بواسطة: claude عرض سجل التدقيق →

درجة الجودة

الهندسة المعمارية

100

قابلية الصيانة

المحتوى

المجتمع

100

الأمان

الامتثال للمواصفات

ماذا يمكنك بناءه

Train ADME Prediction Models

Load Caco-2 permeability data with scaffold splits, train molecular property predictors, and evaluate with standard metrics.

Evaluate Toxicity Predictors

Access hERG, AMES, and DILI toxicity datasets with benchmark protocols to validate safety prediction models.

Generate Novel Drug Candidates

Use molecular oracles like GSK3B and DRD2 to guide generative models toward compounds with desired biological activity.

جرّب هذه الموجهات

Load ADME Dataset

Help me load the Caco2_Wang dataset from TDC with scaffold splitting for training an intestinal permeability predictor.

Run Benchmark Evaluation

Show me how to evaluate my ADME model using the TDC benchmark group with the required 5-seed protocol.

Use Molecular Oracles

I want to evaluate generated SMILES strings using TDC oracles for QED, SA, and GSK3B properties. Show me the workflow.

Drug-Target Interaction Modeling

Load the BindingDB_Kd dataset with cold-drug splitting to ensure my model generalizes to unseen drug compounds.

أفضل الممارسات

Use scaffold splits instead of random splits for realistic model evaluation on novel chemical scaffolds
Run benchmark evaluations with all 5 required seeds to report mean and standard deviation performance
Combine multiple oracles with weighted scoring for multi-objective molecular optimization

تجنب

Avoid random splits for production ADME models as they overestimate performance on similar molecules
Do not report single-seed benchmark results as they may not reflect true model variance
Avoid using oracles as ground truth labels for training since they are predictive models themselves

الأسئلة المتكررة

What datasets are available in PyTDC?

PyTDC includes 60+ datasets covering ADME, toxicity, drug-target interactions, drug-drug interactions, and molecular generation tasks for therapeutic ML.

What is a scaffold split and why use it?

Scaffold splits group molecules by chemical scaffold so test molecules have different core structures than training molecules, simulating real-world generalization.

How do molecular oracles work?

Oracles are pre-trained models that score SMILES strings for properties like drug-likeness (QED), synthetic accessibility (SA), or target binding (GSK3B, DRD2).

What is the 5-seed protocol for benchmarks?

TDC benchmarks require evaluation with 5 different random seeds to compute mean and standard deviation, ensuring robust performance comparisons.

Can I use PyTDC with PyTorch Geometric or DGL?

Yes, TDC provides MolConvert utilities to transform SMILES into PyG graphs, DGL graphs, or other molecular representations like ECFP fingerprints.

What is a cold-drug split for DTI prediction?

Cold-drug splits ensure test set drugs never appear in training, measuring how well models predict binding for completely novel drug compounds.