Q.879 - Molecular replacement and NMR models - what gives ?

In the past two months I have tried to solve four such cases and in two of them the following approach solved them within a couple of hours (using AMORE for the actual MR):

• - align *ALL* NMR models (after removal of any non-protein entities and all hydrogen atoms);
• - remove the obvious outliers;
• - combine the remaining ones into one big composite model and use that *in its entirety* as the probe in the MR (reset all Bs to 20.0 A2 and all occupancies to 1.0; note that some NMR spectroscopists put apocryphal numbers in the B-factor and/or occupancy columns !!!);
• - also create a poly-Ser and a poly-Ala model of the big composite probe (sometimes the poly-Ser works better than the all-atom model; in any case, you would like to get similar solutions for all three models).

The new script file://rose.bmc.uu.se/pub/gerard/omac/multi_probe automates most of this process. All you provide is a PDB file which contains all models without any hetero-entities. The script will split them in the separate models, align each of them with the first model, and print the RMSD and number of aligned CA atoms (all and core). It then asks you if you want to include that model in the probe (you would reject obvious outliers at this stage). The remaining models are concatenated again, Bs and occupancies are reset, and three Molecular Replacement probes are generated (all atoms, poly-Ser and poly-Ala; all of them without hydrogen atoms of course).

Throw these models into AMORE (e.g. with file://rose.bmc.uu.se/pub/gerard/omac/auto_amore.csh) and cross your fingers.

The same approach can also be used if there exist multiple models of your structure in the PDB (either as separate entries or through NCS, or a combination).

One note of caution: the correct solutions are often weak. If you have NCS, you are lucky in that the second molecule should give a much clearer solution than the first.

Checking the correctness of the solution is also easiest if you have NCS: calculate a map with only a poly-Ala model of the solution, average this map (~10 cycles), and check if side chain density shows up in the averaged map.

The refinement tends to start at R ~ 50-55 %, even after rigid- body refinement. This means that you ideally need high-resolution data (unless you prefer to ignore the free R-factor and produce a lousy model, of course .

I recently saw a paper in JMB describing other people's experiences in this area that will be of interest if you are struggling with an (N)MR problem: T. Muller et al., JMB 247, 360-372 (1995).