EYFP mutant fitness prediction
Collection
We trained fitness prediction models using 30K private EYFP mutant fitness data. The fine-tuned model achieves 0.95 spearman's ρ on the test set.
•
4 items
•
Updated
Base model: westlake-repl/SaProt_650M_AF2
Task type: protein-level regression
Dataset: This dataset contains over 100K mutants derived from the wild type EYFP protein. The number of samples for training, validation and test is 100317, 5969 and 5968. 10% of double-site mutants and 10% of triple-site mutants were used for validation and test respectively, and the remains for training. This model was trained by Jia Zheng's lab at Westlake University. The dataset will be released later by this team.
Model input type: Amino acid sequence
Performance (on test set): 0.94 Spearman's ρ
LoRA config:
- r: 8
- lora_dropout: 0.0
- lora_alpha: 16
- target_modules: ["query", "key", "value", "intermediate.dense", "output.dense"]
- modules_to_save: ["classifier"]
Training config:
- optimizer:
- class: AdamW
- betas: (0.9, 0.98)
- weight_decay: 0.01
- learning rate: 1e-4
- epoch: 20
- batch size: 64
- precision: 16-mixed