I ran two ab initio 3D constructions with exactly the same parameters, but they gave me different results. I read an earlier post (Ab-initio job shows different results on repetition) which mentioned that this is caused by different random seed numbers.
This then leads me to wonder how I should know if an ab initio model generated is reliable or correct if there is no homolog structure available for me to compare to?
I would appreciate any advice/insights on this. Thank you very much!
Does your ab initio have high resolution (secondary structure, e.g helices) features, with little smearing or streaking evident? If so this is a good indication that it is likely correct
As @olibclarke suggested, direct validation can only really come from seeing structural elements that are expected/reasonable, or from carrying on with processing and getting a reasonable gold-standard FSC in refinement.
Repetition (i.e. just clone and run the same job over again) is helpful as well. Other helpful diagnostics are to see if the orientation distribution in ab-initio is reasonable - if you only have one or two populated viewing directions, it’s likely that the ab-initio structure will not be correct (since there just isn’t enough data).
You can also try running ab-initio with multiple classes. You should see that at least one class in this case resolves the same/similar structure to what you got with one class, and the other classes may contain junk particles.
The most reliable way may be to perform tomography and sub-tomogram averaging on your grid. This will provide you not only with a valid initial model, but also show you where your proteins are in the ice. We show a case-study in Figure 9 here: