[FEAT] Adding Deep Learning models for similarity search - Githubissues

georgeamccarthy / protein_search

The neural search engine for proteins.

GNU Affero General Public License v3.0

15 stars 6 forks source link

[FEAT] Adding Deep Learning models for similarity search #59

Open fissoreg opened 3 years ago

fissoreg commented 3 years ago

Description:

We are currently using the ProtBERT model, but other models could be used. It would be nice to have a simple selector (dropdown list?) on the frontend to select what model to use to compute the similarity searches.

Why should this be implemented?

ProtBERT has some limitations (i.e. it can't deal with sequences longer then 512 aminoacids, and it's common to have longer protein sequences).
It would be nice to be able to compare the performance of different models.

Additional info

Main models for protein sequences processing: https://github.com/agemagician/ProtTrans
Huggingface models (easy to use and integrate in this project): https://huggingface.co/Rostlab/prot_bert