Final validation of protein binder design using AlphaFold3.

[Mistletoe-git]

Dear AF3 team,

I currently have some questions regarding binder design for certain protein targets. Specifically, after generating binders for protein targets with already resolved structures using some tools, I submit the designed complex sequence to AF3 for final validation. During the AF3 input process, since the binder is designed de novo, there's a high probability that similar proteins don't exist in nature. Therefore, I choose not to provide any MSA or template information to the binder as input. Furthermore, since the protein target exists in nature, we provide its MSA and template. It seems that the currently locally deployed AF3 supports such an asymmetric input method. Like this:

{
  "name": "Binder_Design",
  "modelSeeds": [1],
  "sequences": [
    {
      "protein": {
        "id": "A",
        "sequence": "TARGET_SEQUENCE" 
      }
    },
    {
      "protein": {
        "id": "B",
        "sequence": "BINDER_SEQUENCE",  
        "unpairedMsa": "",
        "pairedMsa": "",
        "templates": []
      }
    }
  ],
  "dialect": "alphafold3",
  "version": 1
}

I'm wondering if this approach is reasonable? Under such circumstances, can the confidence score output by AF3 still be used as screening indicators for design? Or is it that it depends on the specific circumstances? I would appreciate your suggestions!