Is it possible to get dG value(Joules/mole) from rosetta score?
Sort of. Rosetta doesn't really work in laboratory energy units, using an arbitrary energy unit instead.
That said, there has been some work to relate Rosetta energies into laboratory energies, specifically in the context of dG of point mutant stabilities. See Kellogg et al. Proteins. 2011 79(3):830-8 http://dx.doi.org/10.1002/prot.22921 for more.
Say, if i do ligand-docking with many ligands (with the same receptor). Then on what basis i should decide that a particular ligand is the best binding partner for the receptor?
I will be happy if you could say the stepwise method on how to decide! (i.e which scores are to be used and how)
The typical process for ligand docking is to first throw out all those "docked" conformation where the ligand isn't touching the protein. The ligand docking application should give this as a scoreline by default - but they're typically few in number, so it's probably not critical. The next step is to find the lowest energy 5% or so by total energy (ligand+protein), as reported in Rosetta Energy Units (REU). This throws out the wildly improbably conformations. The structures that remain are then ranked by ligand interface energy, basically the calculated energy (in REU) of interaction between the ligand and the protein. The lowest energy structure is considered to be the structure that most likely represents the bound conformation, and the corresponding energy (in REU) represents the relative binding energy of the ligand to the protein. If you take a look at the RosettaLiagand papers by Ian Davis and David Baker they do some benchmarking and show that this quantity is correlated to the experimental binding energy.