someone who uses rosetta ligand docking protocol was asking why score12 was used for ligand docking and not the most recent scoring function which i believe is talaris2014? is it just because it has not been tested yet? or for some other reason? thanks.
First off, don't use score12 itself with ligand docking. There's no reason to do that.
The question then becomes "should I use the (score12-based/pre-talaris) ligand docking score function, or should I use talaris2014, the most recent default?"
To some extent it's personal preference (run some tests and see which works best on your system), but the edge at the moment goes with sticking with the older ligand docking scorefunction. Part of that is lack of comprehensive testing, but mainly it's because when we have tested it, the older ligand docking scorefunction does (slightly) better at predicting the binding conformation of the ligand. (talaris2014 does much better at it than plain score12, though.)
This is due in large part because the ligand docking scorefunction is optimized specifically for ligand docking. When doing ligand position sampling, the interactions which are critical are a bit different than they are when packing a protein. Talaris2014 was optimized with an eye to protein packing/design and protein-protein interactions, with much less attention paid to protein-small molecule interactions. For example, it looks like the electrostatic component of the interaction needs to be upweighted in talaris2014 in order to get better ligand docking results. (talaris2014 has better representation of protein-small molecule electrostatics than plain score12, but not by quite enough.)
At this point I should mention that's there's currently an energy function reoptimization underway (the "beta" effort), which *does* significantly increase the electrostatic interaction component. This reoptimization is explicitly taking protein-small molecule interactions into account, including performance in ligand docking, so I surmise that when it's ready to go as the default scorefunction, it will be the recommended default for everything, including ligand docking. (Though it's not quite ready yet.)
Finally, the recommendation for the ligand docking scorefunction in the pre-talaris-behavior context only really applies to ligand docking proper: that is, trying to find the position of a ligand in a protein binding pocket. If you're doing ligand or enzyme design (finding the best residues to bind to a semi-fixed conformation of a ligand), then it looks like talaris2014 works as well or better than the alternatives. It's basically the side on which you're sampling. If you're primarily sampling the ligand position, use the ligand docking scorefunction. If you're primarily sampling the protein side, use talaris2014.