How to detokenize Protein Embedding Method

user_11092910 0
Viewed 149

I'm using ProtTransBertBFDEmbedder embedding to covert my sequence into embedding format. It returns me an array (length: 1024). My purpose is how can I find the original sequence again by using this 1024-length array. So how can I detokenize/reverse it?

!pip3 install -U bio_embeddings[all] > /dev/null

from bio_embeddings.embed import ProtTransBertBFDEmbedder

embedder_bertbfd = ProtTransBertBFDEmbedder()

embedding = embedder_bertbfd.embed("YSPNNIQHFHEEHLVHFVL")
reduce_per_protein = embedder_bertbfd.reduce_per_protein(embedding)

print(reduce_per_protein)


print(reduce_per_protein.shape)

Output (1024,)

How can I get this original sequence (YSPNNIQHFHEEHLVHFVL) again by using reduce_per_protein

You can use this Original Colab to try

biopython embedding python
1 Answers

You can use BasicAnnotationExtractor() from bio_embeddings.extract module to achieve the desired result as follows:

from bio_embeddings.extract import BasicAnnotationExtractor

print(BasicAnnotationExtractor(reduced_embedding))

Also, you must notice that you need to define your custom model architecture. If not you may run into some issues with its execution.

user_138197140
Related