You will need a really good model for that.

Look at your frames and see if you have splitting of reflections in a few frames 

I just went through all frames but it seems normal to me.

Can you tell me how I can tell splitting from reflection patterns?

I also, just got a MR solution containing one molecule in the ASU where Mathews coefficient predicts there should be 4 (if P64 2 2 space group is correct). This solution lacks number of molecules but the molecule that was placed fit in to electron density very well. 

Should I use this as my new search model? Is there a way to enforce phaser-MR to find more molecules? 

If your true point group is hexagonal you do not expect splitting of reflections (see .jpg) and your twinning is merohedral  (that means "good news")

if your true point group is trigonal or lower you would deal with pseudo-merohedral twinning, which is not good. Those diffraction patterns typically show a few split reflections because the twin operator is not perfectly aligned with the lattice operator. The concept and exceptions thereof is explained here:

http://reference.iucr.org/dictionary/Twin_obliquity

So, presence of splitting is a good indication of pseudo merohedry however the absence is unfortunately no proof against it.

About the solution, well, I would analyze the data a bit more, e.g. with ccp4 instead of just phenix. also try this one:

http://services.mbi.ucla.edu/Twinning/

You could also try mosflm or XDS to see if you can get more reflections off your frames and then run pointless/aimless.

You can "solve" a structure in the wrong space group, however you can't solve a structure if you work in a space group that is too high, e.g. you assign the twin operator to be a crystal symmetry operator.

For MR, if you think it might be twinned you shouldn't use space group P64 2 2 but rather P64 or P62. If you use P3 you are more likely to exclude the twin operator but you will also have to find more molecules, which is difficult when the models are not excellent.

For the tNCS, I think phaser interprets large peaks in the Patterson map as tNCS however there are other reasons those large peaks. For instance, twinning can cause the Patterson peaks in the absence of tNCS if your protein itself has internal symmetry (e.g here 4ZWQ).

Thank you for your generous and detailed explanation.

 I looked through frames and I don't think I have splitting. Just to make sure I attached a screen shot of one of the frames.

I'm trying the approaches that you suggested. 

Thank you again for your expertise and I'll update this post if I make progress. 

Se-Young,

A general comment on the diffraction pattern: The detector is too close to your sample in this experiment. Ideally you want the weakest (~1 I/s) reflection just a few millimeters next to the edge of the detector. That is almost(*) always true for you will observe your reflections better. You can sample larger oscillation angles and hence will be less vulnerable to  radiation damage, you will have more fully observed reflections, less overlapping reflections, and in your case, a better chance of spotting split reflections. 

For your data, keep in mind that you have no evidence yet that your data is twinned. In case of twinning you should find two (or more) good solutions that are mutually exclusive but have similar scores. Lack of an initial MR solution is an indicator for twinning but that is also common for tNCS or for mistakes in data collection and processing. So, you should also try to let phaser correct for tNCS because that would be a great idea if you have tNCS. If you have enough data and are confident about your model, try P1. 

Since your assumption of ASU content was 4 proteins: If your protein is known to form oligomers and your model has a known oligomeric state you could try to use the dimer/tetramer as MR model and look for 1-2 copies because the multimerization is often conserved. This is especially useful if you sample with MR through lower space groups with a higher number of protein chains that need to be found. If you suspect twinning and know your sample well this is exactly the type of knowledge you want to exploit. I think phaser would turn off tNCS mode itself if it realizes that the cell would be too full if it places another tetramer but you should probably monitor that.

S.

(*) I don't know of an example where it would be beneficiary to leave empty space on the detector.