Right, you can only get the ligand_auto_rms_no_super (and thus plot a docking funnel) for those complexes where you have a known native. For figuring out what sort of parameters you need to run (how many structures to generate, etc.) it’s good to have positive controls of similar proteins with similar ligands, run those positive controls through the protocol and confirm that under the conditions you’re using you’re able to adequately recapitulate the appropriate results for the positive controls.
The total_score is more-or-less a measure of the (free) energy of the entire system. So if you have two similar structures, Rosetta predicts that it is more likely to be in the one with the lower total_score. (As with free energies, the absolute numbers aren’t all that meaningful; it’s more the difference between two states that’s relevant.) The interface_delta, on the other hand, specifically measures the energy of association of the ligand with the protein. It’s a measure corresponding (crudely) to the delta-G of the ligand-protein association reaction. So the protein-ligand complex will most likely be found (by Rosetta’s estimation) in those conformations which have low total_scores. Once you’ve identified those most probable conformations, you can gauge the strength of the ligand binding by looking at the interface_delta.
I’m not sure why Rosetta is predicting association with only one of the amino acids, rather than all three of the ones Autodock predicted. More interactions should be favored over fewer interactions, so my off-the-cuff guess is that for the two other residues Rosetta can find more favorable interactions with other parts of the protein, instead of with the ligand. As mentioned above, Rosetta really isn’t currently parametrized well for phosphate binding. What you’re seeing may be a result of that. My guess is that Autodock, being somewhat pharmaceutically biased, is going to be relatively better parametrized on phospho-compounds and phospho-binding proteins. Frankly speaking, if you like the structures and the binding energies coming out of Autodock, I’d go with them. This is not really a situation where Rosetta has a leg up on Autodock.