How wumaotegan/tkam_bio works
TKAM retrieves mechanistically relevant PubChem assays for HepTox and ReproTox, curates their activity data and PFAS molecular inputs, transfers bioactivity learning through a staged multi-task curriculum, and exposes endpoint-specific prediction notebooks. The README documents major workflow stages and integrations, while sampled source confirms KG construction, assay downloading, and multi-task training details; notebook internals and some integrations are not sampled, so uncertain wiring is omitted.
Main entry points: workflow_hepatotoxicity_bioassay_retrieval.ipynb HepTox retrieval · workflow_reproductive_toxicity_bioassay_retrieval.ipynb ReproTox retrieval · data Assay and toxicity data · get_info_pub.py Assay download
Components 19
Bioassay retrieval
- KG and embeddings
code/bioassay_retrieval/generate_knowledge_graph_qwen3_base.py - Bioassay knowledgebase
data - HepTox retrieval
code/bioassay_retrieval/workflow_hepatotoxicity_bioassay_retrieval.ipynb - ReproTox retrieval
code/bioassay_retrieval/workflow_reproductive_toxicity_bioassay_retrieval.ipynb - Assay download
code/bioassay_retrieval/get_info_pub.py
Curation and representation
- Assay curation
code/bioassay_retrieval - SMILES standardization
code/chembert_pfas/PFAS_SMILES_standardization.ipynb - PFAS encoder
code/chembert_pfas/ChemBERTa-100M_full_finetune.ipynb
Curriculum training
- Multi-task curriculum
code/multi_tasks_training/multitask-pretraining-with-pruning-FT_heptox.ipynb - In-vivo trainer
code/multi_tasks_training/invivo_train_curriculum_save_model.py - Assay and toxicity data
data
Prediction
- Endpoint models
code/predict/predict_models - HepTox prediction
code/predict/predict_heptox.ipynb - ReproTox prediction
code/predict/predict_reprotox.ipynb
Other components
- Researcher
- AOP-Wiki
- PubChem
- Model services
- Assay descriptions
data
Connections 28
- Researcher → HepTox retrieval: runs workflow
- Researcher → ReproTox retrieval: runs workflow
- Assay descriptions → KG and embeddings: provides descriptions
- KG and embeddings → Model services: requests extraction and embeddings
- KG and embeddings → Bioassay knowledgebase: builds graph
- HepTox retrieval → AOP-Wiki: queries AOPs
- ReproTox retrieval → AOP-Wiki: queries AOPs
- HepTox retrieval → Bioassay knowledgebase: retrieves and expands assays
- ReproTox retrieval → Bioassay knowledgebase: retrieves and expands assays
- HepTox retrieval → Model services: reranks and reasons
- ReproTox retrieval → Model services: reranks and reasons
- HepTox retrieval → Assay download: downloads selected assays
- ReproTox retrieval → Assay download: downloads selected assays
- Assay download → PubChem: fetches activity records
- Assay download → Assay curation: supplies raw records
- Assay curation → Assay and toxicity data: writes curated tasks
- SMILES standardization → Assay and toxicity data: annotates PFAS inputs
- Assay and toxicity data → PFAS encoder: supplies PFAS SMILES
- PFAS encoder → Multi-task curriculum: provides adapted encoder
- Assay and toxicity data → Multi-task curriculum: supplies assay tasks
- Multi-task curriculum → In-vivo trainer: passes trained stages
- Assay and toxicity data → In-vivo trainer: supplies in-vivo labels
- In-vivo trainer → Endpoint models: saves models
- Multi-task curriculum → Endpoint models: produces model artifacts
- Researcher → HepTox prediction: submits chemicals
- Researcher → ReproTox prediction: submits chemicals
- HepTox prediction → Endpoint models: loads endpoint model
- ReproTox prediction → Endpoint models: loads endpoint model