Modeling with cryo-EM


Step 5: Assess the quality of the models

In the previous step, 5 models were built for the target sequence. However, how can the "best" model be selected? There are various methods provided in MODELLER for so-called model assessment. Perhaps the simplest is the molecular PDF (molpdf), which is the standard MODELLER scoring function, and is simply the sum of all restraints. The molpdf is written into a REMARK in each generated PDB file. Additional assessment methods include DOPE and GA341, and are only calculated by AutoModel if you list them in assess_methods (although you can go back and recalculate them for a given PDB file later). The molpdf and DOPE scores are not 'absolute' measures, in the sense that they can only be used to rank models (molpdf is specific for a given set of restraints, and DOPE for a given target sequence). Other scores are transferable. For example GA341 scores always range from 0.0 (worst) to 1.0 (native-like); however GA341 is not as good as DOPE at distinguishing 'good' models from 'bad' models.

The MODELLER log file from an AutoModel run includes a summary at the end of all models built, and their assessment scores. That from the previous step is shown below as an example. (Note, however, that the actual values and generated PDB files may differ on other computers, due to slight differences in floating point representation and rounding on these systems. This is nothing to worry about.)

>> Summary of successfully produced models:
Filename                          molpdf     DOPE score
-------------------------------------------------------
TvLDH.B99990001.pdb           1626.69543   -38816.71484
TvLDH.B99990002.pdb           1689.35669   -38618.24609
TvLDH.B99990003.pdb           1631.80737   -38550.80078
TvLDH.B99990004.pdb           1798.11438   -38452.23047
TvLDH.B99990005.pdb           1711.15381   -38474.54688

Excerpts of the file model-single.log

These data would suggest that all of the models are reasonably good, since the scores are close, but TvLDH.B99990001.pdb could be considered the best due to it having both the lowest molpdf and the lowest DOPE score.

On to the next step, or back to the index.