search

microsoft / Graphormer

Graphormer is a general-purpose deep learning backbone for molecular modeling.
MIT License
2.16k stars 337 forks source link

DiG protein explanation #202

Open bio-rat opened 2 months ago

bio-rat commented 2 months ago

Hi team,

First of all thank you for making this wonderful program public. I was testing out the program today for the protein part of DiG and followed with 1 of the 6 provided examples. I have two questions:

  1. How to use the output: it has two .npz files one for init and one for final. What should I do with them.
  2. I want to use the program for a pdb that is not in the provided dataset. How would I go with generating the input .pkl files?

It would be really helpful for us biologists if you could give a short tutorial on how to use the program!

Please have a great week!

Best

dcbiton commented 1 month ago

Hi @bio-rat , I am not from the Microsoft team but I was able to implement the protein model using my own pdb and generate protein conformations. The input .pkl files are from the output of the evoformer of alphafold2. For my case, I had the implementation of evoformer from openfold and I saved the pkl files from the 'single' and 'pair' embeddings from the output of the iteration of evoformer The openfold needs the (1) .fasta file and (2) mmcif dir and (3) .a3m alignment. But you can get these from the actual implementation of alphafold2. I hope this helps