---
base_model: westlake-repl/SaProt_35M_AF2
library_name: peft
license: mit
---
# Base model: [westlake-repl/SaProt_35M_AF2](https://huggingface.co/westlake-repl/SaProt_35M_AF2)

# Model Card for Model ID

<!-- Provide a quick summary of what the model is/does. -->
This model is trained on a sigle site deep mutation scanning dataset and
can be used to predict fitness score of mutant amino acid sequence of protein [BLAT_ECOLX](https://www.uniprot.org/uniprotkb/P62593/entry) (Beta-lactamase TEM). 

## Protein Function
TEM-type are the most prevalent beta-lactamases in enterobacteria; 
they hydrolyze the beta-lactam bond in susceptible beta-lactam antibiotics, thus conferring resistance to penicillins and cephalosporins. 

### Task type
protein level regression

### Dataset description
The dataset is from [Deep generative models of genetic variation capture the effects of mutations](https://www.nature.com/articles/s41592-018-0138-4).
And can also be found on [SaprotHub dataset](https://huggingface.co/datasets/SaProtHub/DMS_BLAT_ECOLX).

Label means fitness score of each mutant amino acid sequence, ranging from 0 to positive infinity where 1 is the value of wildtype,
larger than 1 means higher fitness than wildtype.

### Model input type
Amino acid sequence

### Performance
0.80 Spearman's ρ

### LoRA config
lora_dropout: 0.0

lora_alpha: 16

target_modules: ["query", "key", "value", "intermediate.dense", "output.dense"]

modules_to_save: ["classifier"]

### Training config
class: AdamW

betas: (0.9, 0.98)

weight_decay: 0.01

learning rate: 1e-4

epoch: 50

batch size: 64

precision: 16-mixed