HADDOCK itself is capable to treat chemical shift perturbation data (NMR titration data). These are so-called AIRs (Ambiguous Interaction Restraints). Presumably, protein signals should be assigned before this.

Alternative way of the validation of the correctness of the calculated protein-ligand complex can be via generation the list of protein residues surrounding the ligand (typically within 4Å from the ligand). Such list should resemble (not necessary to be identical) list of the residues having substantial chemical shift changes in NMR titration experiments.

Haddock encodes biochemical/biophysical data of the complex in terms of ambiguous intermolecular distance restrains to drive the docking.

Although they provide valuable information about the interface of the two proteins, they also lack of orientation content. In general, these intermolecular distance restraints are rough and ambiguous, and there are many false solutions (minima) that agree with them.

To validate the docked complex, you could point - mutate residues that were predicted by the generated model to be involved in the interaction. Then check if they perturb the binding equilibrium.

De Vries, S. J., Melquiond, A. S. J., Kastritis, P. L., Karaca, E., Bordogna, A., Van Dijk, M., Rodrigues, J. P. G. L. M., et al. (2010). Strengths and weaknesses of data-driven docking in critical assessment of prediction of interactions. Proteins, 78(15), 3242–3249.

You could also used methods such as SAXS to asses the quality of the complex.

Cordeiro, T. N., Schmidt, H., Madrid, C., Juárez, A., Bernadó, P., Griesinger, C., García, J., et al. (2011). Indirect DNA Readout by an H-NS Related Protein: Structure of the DNA Complex of the C-Terminal Domain of Ler. PLoS pathogens, 7(11), e1002380. doi:10.1371/journal.ppat.1002380

Given SAXS experimental data, CRYSOL can fit the theoretical scattering curve by minimizing the discrepancy (reduced chi-squared). If the theoretical structure agrees with SAXS data, this does not yet prove it uniqueness at high-resolution, but any model failing to fit the SAXS data definitely doesn’t represent the average macromolecular conformation in solution. Therefore, SAXS is an excellent validation method for high-resolution structures and together with other techniques and computational tools is a powerful tool for tackling challenging and relevant biological problems. SAXS can significantly improved accuracy in solution structure determination.

Thanking a lot to Vladimir Polshakov , Tiago Cordeiro for your valuable comments !