3D Flex Training dependent on previous refinement job? Minimum size for 3D Flex?

3D Flexible Refinement
1

Hi everyone!

I am playing around with 3D Flex Refinement, and I am wondering if the training result is dependent on the preceeding Refinement jobs (NU-Refine, Homo refine…)? From this forum, I know that 3D Flex benefits from CTF refinements as discussed here . Apart from refining CTF, should I try to push the resolution to as high as possible before running 3D Flex, since the training does use a downsampled refined map from my understanding.

Appreciate any thoughts on this!
Khoa

2

Hi @lecongmi001! In general, we expect that 3D Flex will benefit most from having a clean input particle stack. Additionally, if you have the largest rigid part of the molecule well-aligned in the input particles the model will have to use less of the latent space to model rigid-body shifts of the entire particle, which may improve the results as well.