Pyrosage: Environmental & Toxicity GNN Models

Pyrosage is a suite of 24 Graph Neural Network models (AttentiveFP, PyTorch Geometric) for predicting environmental and toxicity properties directly from SMILES. Models use enhanced molecular graphs (10 atom features, 6 bond features) for robust performance.

Models

Classification (7)

• AMES (mutagenicity) — AMES.csv
• Endocrine Disruption
  • NR-AR.csv
  • NR-AhR.csv
  • NR-ER.csv
  • NR-aromatase.csv
• Eye hazards
  • Eye_Corrosion.csv
  • Eye_Irritation.csv

Regression (17)

• Partitioning:
  • KOA.csv
  • KOC.csv
  • KOW.csv
• Solubility & volatility:
  • SW.csv
  • KH.csv
  • PLV.csv
• Reactivity & fate:
  • kAOH.csv
  • tbiodeg.csv
• Toxicity:
  • LC50.csv
  • LD50_Zhu.csv
  • TBP.csv
  • tfishbio.csv
  • TMP.csv
• Acid/base:
  • pKa_acidic.csv
  • pKa_basic.csv

Additional endpoints

• FBA.csv
• FBC.csv

Usage

• Provide a SMILES string and select a model.
• Outputs: probability/logit for classification; scalar value for regression.

Models are bundled in a single service for efficient access and consistent preprocessing.