search

chaidiscovery / chai-lab

Chai-1, SOTA model for biomolecular structure prediction
https://www.chaidiscovery.com
Other
1.28k stars 159 forks source link

Request for Integration of pAE (Predicted Aligned Error) Score in Chai-1 Outputs #20

Open assiduous006 opened 2 months ago

assiduous006 commented 2 months ago

Dear Chai Discovery Team,

First of all, I want to express my gratitude for making the Chai-1 codebase open-source and accessible to the community. Your work is truly invaluable and has great potential to push forward advancements in protein structure prediction.

I am writing to kindly request the inclusion of the Predicted Aligned Error (pAE) score in the outputs of the Chai-1 model. The pAE score has proven to be highly useful for assessing the reliability of various regions in protein structure predictions, and its availability could further enhance the utility of the model for downstream tasks like structural analysis and drug discovery.

Could you please consider integrating the pAE score into the output of Chai-1? I believe that this would greatly benefit users, particularly those relying on confidence measures for their research.

Thank you very much for your time, and for your continued contributions to the field. I look forward to any thoughts or feedback you may have on this request.

Warm regards, Eric

GXcells commented 2 months ago

Hi, not pAE but other metrics were implemented today #4 But for me I had to install chai_lab manually using 

git clone https://github.com/chaidiscovery/chai-lab/
cd chai-lab
pip install .

Because pip install chai_lab was not pulling the last commits

jayunruh commented 2 months ago

I would like to upvote this issue. pAE outputs are crucial for quantitative comparison against other prediction models. In my own install of this, I achieved this goal by converting the confidence_scores output (typically ignored in run_inference) from run_folding_on_context and converting it to a dictionary with:

confdict = {
    'pae': confidence_scores.pae.cpu().numpy(),
    'plddt': confidence_scores.pae.cpu().numpy()
}

I then added that to the return of run_inference. Fairly easy fix.

stianale commented 1 month ago

I would like to upvote this issue. pAE outputs are crucial for quantitative comparison against other prediction models. In my own install of this, I achieved this goal by converting the confidence_scores output (typically ignored in run_inference) from run_folding_on_context and converting it to a dictionary with:

confdict = {
    'pae': confidence_scores.pae.cpu().numpy(),
    'plddt': confidence_scores.pae.cpu().numpy()
}

I then added that to the return of run_inference. Fairly easy fix.

Can this be applied to already run analyses and score files, or must one run analyses anew to get PAE?