README.md

## ProtBert-BDF-IS

### Model Description

ProtBert-BFD-IS is a a model fine-tuned on the pre-trained ProtBert-BFD model for the purpose of sequence classification. It takes a protein sequence input and predicts whether the protein is soluble or insoluble.
ProtBert-BFD-IS has been fine-tuned using 3 different training datasets.

**Finetuned from model:** Rostlab/prot_bert_bfd

GitHub repository with relevant files: https://github.com/VitaRin/ProtBert-IS

## Uses

It can be directly used with the pipeline on singular sequences:

```
from transformers import BertModel, BertTokenizer
import re

pipeline = TextClassificationPipeline(
    model=AutoModelForSequenceClassification.from_pretrained("VitaRin/ProtBert-IS"),
    tokenizer=AutoTokenizer.from_pretrained("VitaRin/ProtBert-IS"),
    device=0
)
sequence = "A E T C Z A O"
sequence = re.sub(r"[UZOB]", "X", sequence)
output = pipeline(sequence)
```

Or read multiple sequences from a .fasta file:

```from transformers import AutoTokenizer, AutoModelForSequenceClassification, TextClassificationPipeline
import re

pipeline = TextClassificationPipeline(
    model=AutoModelForSequenceClassification.from_pretrained("VitaRin/ProtBert-IS"),
    tokenizer=AutoTokenizer.from_pretrained("VitaRin/ProtBert-IS"),
    device=0
)

with open("input.fasta", "r") as f:
    data = f.read().split(">")

data.remove(data[0])
sequences = []

for d in data:
    d = d.split('\n', 1)[-1].replace('\n', '').replace('', ' ')
    sequences.append(d)

sequences = [re.sub(r"[UZOB]", "X", sequence) for sequence in sequences]
print(pipeline(sequences))
```