torchdrug

PyTorch-native graph neural networks for molecules and proteins. Use when building custom GNN architectures for drug discovery, protein modeling, or knowledge graph reasoning. Best for custom model development, protein property prediction, retrosynthesis. For pre-trained models and diverse featurizers use deepchem; for benchmark datasets use pytdc.

1.18x

Quality

71%

Does it follow best practices?

Impact

94%

1.18x

Average score across 9 eval scenarios

Securityby

Advisory

Suggest reviewing before use

Optimize this skill with Tessl

npx tessl skill review --optimize ./scientific-skills/torchdrug/SKILL.md

Evaluation results

100%

16%

CNS Drug Candidate Screening Pipeline

Molecular GNN model selection and training

Criteria

Without context

With context

GIN model choice

100%

Scaffold split

100%

Node feature dim from dataset

100%

Edge feature dim from dataset

100%

Batch norm enabled

100%

BCE criterion

100%

AUROC metric

100%

AUPRC metric

100%

PropertyPrediction task

100%

uv pip install

100%

98%

20%

Biomedical Knowledge Graph Drug Repurposing

Knowledge graph completion for drug repurposing

Criteria

Without context

With context

RotatE or ComplEx model

100%

Hetionet dataset

100%

KnowledgeGraphCompletion task

100%

adversarial_temperature param

100%

num_negative param

100%

num_entity from dataset

100%

num_relation from dataset

100%

BCE criterion for KG

100%

uv pip install

Compound-disease query

100%

32%

Automated Synthesis Route Prediction

Retrosynthesis two-stage pipeline

Criteria

Without context

With context

USPTO50k dataset

100%

RGCN for center identification

100%

GIN for synthon completion

100%

CenterIdentification task

100%

SynthonCompletion task

100%

num_bond_type for RGCN

100%

Two separate models

100%

node_feature_dim from dataset

100%

uv pip install

100%

top_k for center identification

100%

90%

10%

Protein Thermostability Prediction from Sequence

Protein sequence model selection and fine-tuning

Criteria

Without context

With context

ESM model selected

100%

Fluorescence dataset

100%

PropertyPrediction task

100%

MSE criterion

100%

MAE metric

100%

RMSE metric

100%

Small fine-tuning learning rate

100%

uv pip install

No structure-based imports

100%

92%

16%

De Novo Drug Candidate Generation for a CNS Target

Property-optimized molecular generation with GCPN

Criteria

Without context

With context

GCPN task

100%

GIN backbone for GCPN

100%

ZINC dataset

100%

Multi-objective reward

100%

RDKit validity check

100%

PPO criterion

100%

uv pip install

No GraphAutoregressiveFlow

100%

88%

Protein Fold Classification from 3D Structure

Protein structure-based function prediction with GearNet

Criteria

Without context

With context

GearNet model selected

100%

Fold dataset

83%

100%

Multiple edge types

100%

50%

data.Protein.from_pdb usage

100%

CE criterion

100%

PropertyPrediction task

100%

No ESM or sequence-only model

100%

dataset.node_feature_dim for input

100%

uv pip install

Repository: K-Dense-AI/claude-scientific-skills
Commit: 25e1c0f

Evaluated: about 2 months ago
Agent: Claude Code
Model: Claude Sonnet 4.6

Table of Contents

CNS Drug Candidate Screening Pipeline Biomedical Knowledge Graph Drug Repurposing Automated Synthesis Route Prediction Protein Thermostability Prediction from Sequence De Novo Drug Candidate Generation for a CNS Target Protein Fold Classification from 3D Structure

Is this your skill?

If you maintain this skill, you can claim it as your own. Once claimed, you can manage eval scenarios, bundle related skills, attach documentation or rules, and ensure cross-agent compatibility.