Using Pairformer as a Standalone Module for MSA Embedding

[real-junjiezhang]

Hello,

Thanks for your great work in advance!

I'm interested in using the Pairformer component independently from the full Protenix pipeline. Specifically, I would like to know if it's possible to utilize only the Pairformer part where:

Input: An MSA (a3m file)
Output: The embedding of the MSA produced by Pairformer

My plan is to integrate Pairformer as an independent module in my code and retrain just this part. Could you please advise if this approach is feasible? Are there any inherent dependencies or modifications required to decouple Pairformer from the rest of the system?

Thank you for your help and looking forward to your guidance!

Bests

[zhangyuxuann]

@real-junjiezhang Pairformer is independent from the rest of the system. In this repo, you can use the get_pairformer_output to get the embedding of pairformer.

Protenix/protenix/model/protenix.py

Line 115 in 9765426

def get_pairformer_output(

However, if you only input an a3m file, you need to do some development to convert some features to adapt to the args of pairformer.