cmartell/Model-75C_Aggregation-650M

Base Model: westlake-repl/SaProt_650M_AF2

Model Card for Model ID

This model predicts protein aggregation induced by 75°C stress.

Task type

Protein level regression

Dataset description

• Trained on experimental data from Global Analysis of Aggregation Determinants in Small Protein Domains The dataset can also be found on SaprotHub dataset.

Model input type:

SA (Structure-aware) protein sequence

Training Details

Performance (on test set)

Spearman's ρ = 0.75
R² = 0.55

LoRA Config:

• r: 8
• lora_dropout: 0.1
• lora_alpha: 16
• target_modules: ["query", "intermediate.dense", "value", "output.dense", "key"]
• modules_to_save: ["classifier"]

Training Config

• Optimizer:
  • class: AdamW
  • betas: (0.9,0.98)
  • weight decay: 0.01
• learning rate: 5e-4
• batch size: 1
• gradient accumulation steps: 64
• effective batch size: 64
• epochs: 35
• precision: 16-mixed