I have used 3D Flex on my system and the preliminary results are very promising. My complex can be abstracted as 3 domains (A, B, C), where A and B are both connected to C. A and B are in close proximity and even loosely interact, but they also appear to move quite a bit against each other (I have also done 3DVA before). Therefore I created custom segmentation and specified their connectivity (C>B,C>A). As expected, 3D Flex learned movement dominated by movement of B and C relative to each other. However, the reconstructed half-maps appear to be severely affected by the mesh segmentation: At the point where the two domains come into contact, the map has a sharp, stepped edge which looks like it stems from the 3D flex grid.
Half map after 3D reconstruction
How can I remedy this? Should I use a finer mesh for training?
The FSC of my final map appears to be between 3.3 and 3.5A, I restricted the resolution to 3.9 during training. The closest distance between A and B is ~5A, could that be the problem?
Also, can I use local resolution estimation after 3DFlex? What would be the proper way to go about this? I tried it already, but so far, the sharp-edged artefact makes it pretty non-sensical:
Resolution map in frequency
I would appreciate some help!