search

integrativemodeling / mediator

Modeling of the yeast Mediator complex
https://salilab.org/mediator/
Creative Commons Attribution Share Alike 4.0 International
0 stars 0 forks source link

Cross correlations for EM fits seem low #5

Open benmwebb opened 7 years ago

benmwebb commented 7 years ago

I was just looking at rescoring the deposited RMF files and while the EM scores look reasonably close to what's in the stat files, the cross correlation coefficients seem low (around 0.2). @Pellarin , is this to be expected for Mediator?

e.g. if I rescore the top ranking model for the first cluster, i.e. analysis/clustering/kmeans_weight_500_4/cluster.0/0.rmf3

... I get for the middle module an EM score of 181.99 and a CCC of 0.22, and for the tail an EM score of 217.13 and CCC of 0.17. The stat file contains 'GaussianEMRestraint_None': ' 211.99099168'. No label was assigned, so that must be the tail EM score, and my value is within 2% so I'm confident the coordinates are correct.

@cgreenberg, for CCC I'm reporting what you call cc in the code here: https://github.com/salilab/imp/blob/develop/modules/isd/src/GaussianEMRestraint.cpp#L203 Maybe I did something dumb and that's not really the cross correlation?

benmwebb commented 7 years ago

@cgreenberg confirms this to really be the cross correlation (sort of) and that, given the coarse representation of the system, values this low aren't surprising.

benmwebb commented 7 years ago

BTW, the rescore is script is here. It's just setting up the same representation and restraints as in the original modeling, reading in one of the RMF files, then doing a single score evaluation. Only difference is that I don't call scale_bead_radii(), center_model_on_target_density() or optimize_floppy_bodies() since I don't want to change the coordinates from what's in the RMF.