metadata

license: bigscience-openrail-m
widget:
  - text: >-
      M[MASK]LWMRLLPLLALLALWGPDPAAAFVNQHLCGSHLVEALYLVCGERGFFYTPKTRREAEDLQVGQVELGGGPGAGSLQPLALEGSLQKRGIVEQCCTSICSLYQLENYCN
datasets:
  - Ensembl
pipeline_tag: fill-mask
tags:
  - biology
  - medical

BERT base for proteins

This is bidirectional transformer pretrained on amino-acid sequences of human proteins.

Example: Insulin (P01308)

MALWMRLLPLLALLALWGPDPAAAFVNQHLCGSHLVEALYLVCGERGFFYTPKTRREAEDLQVGQVELGGGPGAGSLQPLALEGSLQKRGIVEQCCTSICSLYQLENYCN

The model was trained using the masked-language-modeling objective.

Intended uses

This model is primarily aimed at being fine-tuned on the following tasks:

protein function
molecule-to-gene-expression mapping
cell targeting

How to use in your code

from transformers import BertTokenizerFast, BertModel
checkpoint = 'unikei/bert-base-proteins'
tokenizer = BertTokenizerFast.from_pretrained(checkpoint)
model = BertModel.from_pretrained(checkpoint)

example = 'MALWMRLLPLLALLALWGPDPAAAFVNQHLCGSHLVEALYLVCGERGFFYTPKTRREAEDLQVGQVELGGGPGAGSLQPLALEGSLQKRGIVEQCCTSICSLYQLENYCN'
tokens = tokenizer(example, return_tensors='pt')
predictions = model(**tokens)