In liganddock documentation, it is said "interface_delta" (in score.sc ) gives estimate of the binding energy.
Can this be related to experimental units of binding energy i.e like in terms of kJ/mole?
Are there any experimental evidence to show that the "interface_delta" correlates with that of the experimental binding energy?
The short answer is that Rosetta energies are all in REU (Rosetta Energy Units), and there isn't a standard conversion from REU to kJ/mole or kcal/mole. (And moreover what an REU is varies based on which weights file. options and protocol you're using.)
That said, there's been a number of benchmarks which show that REU values are correlated with corresponding kcal/mole values. The rule of thumb is that an REU is close to a kcal/mol, but only *very* roughly. For protein-ligand docking, the reference you're looking for would probably be Meiler & Baker Proteins. 2006;65(3):538-48 (http://dx.doi.org/10.1002/prot.21086 - See Figure 4). This gives a conversion factor of about 2 REU to 1 kcal/mol. Note that the algorithm has changed since then, so the numerical conversion factor from that paper might not apply anymore.
If you're interested in getting numeric predictions, my recommendation is to identify a benchmark set of compounds like the one you're trying to predict that have known experimental values and then run the protocol that you intend to use on that benchmark set. You can then get a correlation line from the know examples which can hopefully be applied to your unknown compound.