Hi, running exactly the same 3D refinement job with the same parameters (below) gives final resolution 4.3 A in relion2.0 (and map looks OK), but in relion2.0.6 for the same run there are many more iterations and apparent final resolution is 3.7 A. However, the map looks more like 4.3 A with added noise, it is not really 3.7. It looks as if default tau2_fudge increased to some very high value in 2.0.6, resulting in over-refinement (although tau2fudge is reported as 1 in run_model.star). The same thing happens with either tighter or broader masks during refinement. Two random halves of data are properly defined, this we checked.
Has anybody else seen something similar and what can be done to restore 2.0 behavior in 2.0.6?
Many thanks!
The job parameters are: `which relion_refine_mpi` --o Refine3D/job288/run --auto_refine --split_random_halves --i particles.star --ref Models/c512box_30A.mrc --firstiter_cc --ini_high 30 --dont_combine_weights_via_disc --no_parallel_disc_io --pool 6 --ctf --ctf_corrected_ref --particle_diameter 350 --flatten_solvent --zero_mask --solvent_mask Models/c512box_30A_450thresh_2ext_7soft.mrc --solvent_correct_fsc --oversampling 1 --healpix_order 2 --auto_local_healpix_order 4 --offset_range 20 --offset_step 4 --sym C1 --low_resol_join_halves 40 --norm --scale --j 14
|