Strange FSC curve and reconstruction from 3D FLEX

Hi all, I’m playing around with a datasets in 3dFlex and I’m having problems with the reconstruction step.

The training curves look ok. But the reconstruction FSC is very strange.
The strange increase at high resolution persists no matter which hyper parameters I use and how much I dilate the mask for FSC validation.

The local resolution estimation suggests that the flexible parts of the protein have been resolved to 2.5 A. But visually, the flexible regions have much less detail in the flex model than the conventional reconstruction.

Any ideas what could be going on? This is one of the datasets used in the 3dflex preprint.

Here are the training curves (couldn’t include them in the original post because I’m a new user).

One thing you can try is increasing the number of ‘Max BFGS iterations’. Once I changed that to 40, my dataset FSC looked normal.


Oh that worked! thank you!

@joonpark @Miro-Astore Hi! I would like to ask which kind of machine are you running 3d flex refinement on. As the issue (3D Flex pytorch issue on RTX4090 - #13 by wtempel) exists, I can not run 3d flex in cryosparc v4.2.1 on RTX 4090. I have confirmed that cryosparcw call nvcc --version returns 11.8. But in CryoSPARC v4.2.1, after running “/home/xx/cryosparc/cryosparc_worker/bin/cryosparcw install-3dflex” to install pytorch with CUDA version 11.8, it returns the “1.13.1+cu117”.

Hi @XianniZhong

I apologize for such a late response. I hope the issue has been resolved. If you are asking about the GPU I used, I’ve used both RTX A4000 and RTX 2080 Ti.

Hi! Also apologies for the late reply. these were run on A100s.