Datasets used in the tox21 challenge

With the rise of deep learning models and the successful result showing in different domains (such as Computer vision and Natural language processing)researchers and laboratories of chem-informatics try to apply these techniques in drug design and discovery. recently,the application of Deep Learning in this area of research has made a good progr…

some scripts using deepchem

Binary classification neural networks built from scratch for Tox21 chemical data.

MoltiTox: a multimodal fusion model for molecular toxicity prediction

A Polymer Toxicity Prediction Tool using PSMILE Strings

Multi-label classification problem.

Descriptive analysis and QSAR modelling for tox_21 datasets

ChemAI: Multi-property chemical prediction using Random Forest and Graph Neural Networks

Tox21 quantitative high throughput screening (qHTS) 10K library data

Scaffold-aware GCN with attention-based substructure pooling for multi-task Tox21 toxicity prediction. 12-task model with scaffold-split evaluation achieving 0.789 AUC-ROC on best assay.

EPA data on 1,800 chemicals across 700+ assay endpoints

Native fine-tuning of Mistral-7B on the Tox21 dataset using LoRA and 4-bit quantization, achieving a competitive 0.72 ROC-AUC.