salilab / imp

The Integrative Modeling Platform
https://integrativemodeling.org
GNU General Public License v3.0
74 stars 30 forks source link

Density sampling and EM projections not working as expected #915

Open ichem001 opened 8 years ago

ichem001 commented 8 years ago

The current version of the EM and EM2D modules uses the KernelParameters.h file to sample the density at a given resolution and angstrom per pixel (apix). However both uses the RadiusDependentKernelParameters (RDKP) to define the volume and sigma at which the density will be sampled.

At low resolution, the models looked okay given the resolution. However, upon using high resolution data, the simulated maps (respectively the projection) are not correct. This is easily shown by using standard EM software such as EMAN2.

Disabling the RadiusDependentKernelParameters seems to solve one of the issue at high resolution (proper density location but wrong voxel value even when using NumberOfElectron rather than mass when calculating the KP) but low resolution is still not working as expected. Ubiquitin looks like a ball at 15A losing all the features (EMAN2 actually give proper features even at 15A resolution).

Obviously, disabling the RDKP is not ideal. I am not sure why it was implemented but I am sure there is a good reason for it to be there.

TL/TR: Incorrect density sampling (resolution and voxel intensity) from PDB when using RDKP.

cgreenberg commented 8 years ago

Even though this issue is mostly solved with recent changes in IMP::em, I'm still finding errors in simulated maps. They still have a slight offset from the structure they're simulated from.

ichem001 commented 8 years ago

What do you mean by offset? Resolution-wise or coordinate-wise?

benmwebb commented 8 years ago

If you mean coordinate-wise, it would be simple to write a (failing) test that simulated a map from a single atom, then compared the map centroid to the atom coordinates, no?