Created README.md

README.md

@@ -0,0 +1,55 @@
+
+## ProtBert-IS
+
+### Model Description
+
+ProtBert-IS is a a model fine-tuned on the pre-trained ProtBert model for the purpose of sequence classification. It takes a protein sequence input and predicts whether the protein is soluble or insoluble.
+ProtBert-IS has been fine-tuned using 3 different training datasets.
+
+**Finetuned from model:** Rostlab/prot_bert
+
+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))
+```
+