Solving preferred orientation problem by AI algorithms

Dear community,

I found two recently released paper talking about using self-supervised algorithms to solve the preferred orientation problem:

cryoPROS: Addressing preferred orientation in single-particle cryo-EM through AI-generated auxiliary particles

spIsoNet: Resolving the Preferred Orientation Problem in CryoEM Reconstruction with Self-Supervised Deep Learning

Has anyone tried these ways? or any comments on these algorithms? I have a protein showing a little bit of preferred orientation, and I feel these self-supervised algorithms could be the right solution for us.


Are these available out there now?

1 Like

Not to my knowledge for the first, but an update to spIsoNet has just hit bioRxiv…

And code is here:

Not tested yet.


I am giving that a go and it seems I am stuck at

.conda/envs/spisonet/lib/python3.10/site-packages/torch/cuda/ UserWarning: Can't initialize NVML

I have it running on my system… Error looks like a cuda version or nvidia driver mismatch of some sort, but you might have more luck asking on the spIsoNet github page :slight_smile:

Agree. thanks @olibclarke

1 Like

NVML errors usually mean the system needs a reboot for me.

1 Like

What kind of results have you seen using the software?

Really exciting to see this!
However I got this error:
BlockingIOError: [Errno 11] Resource temporarily unavailable

Does anyone know how to solve this?
Thank you

That did the trick for NVML and then for torch errors i upgraded gcc

1 Like

Ah, I haven’t run gcc 4 in years… good to know to keep an eye out for, cheers!