Here what you could try:

First, make sure your data are not twinned.

Second (assuming your data are not twinned), use Self Rotation function (I like GLRF) to determine the exact number of copies in the asu. A search for k=45 should 'show' the existence and orientation of a 8-fold ncs axis. If you don't see anything at k=45, then you don't have 8 copies in the asu. Try different options (e.g. k=60 for 6-fold, k=90 for 4-fold, etc), ultimately k=180 should give you an equal number of 2-fold axes as the number of particles in the asu.

Third, try MR with whatever 'geometric' model you have (better smaller then larger).

Don't be bothered by low LLG values.If you have a solution that 'makes sense' (packing-wise)

compute self rotation functions with Icalc and compare them to those calculated with Iobs.

The two should look alike, if the solution is correct.

Fourth, assuming the above have worked, you need to go from a 'geometric' model to a

'physical entity' to calculate phases. Mask you protomer and bootstrap real phases

by ncs-averaging density modification (e.g. resolve, dm, etc).

Fifth, calculate an averaged map, build/improve your initial model and used Pavel's new real_space_refinemnet module to refine it against the averaged density.

Real space refinement has larger radius of convergence that classical rigid/positional refinement

and works much better at medium resolution (~4A).

Good luck!

Gino