A difference in HADDOCK scores, RMSD and Z-scores for the identical protein-protein complex run multiple times

I am trying to dock two proteins together. I have carried out several jobs of the same proteins with the identical docking parameters. However, each job’s result gives me a different HADDOCK score, RMSD and Z-score.
Of course, the differences in scores are by +/- 5 but I would still like to know why there is a difference in scores if all the parameters and proteins are the same.

P.S. I am a student, and I am trying to compare binding affinities and energy of different protein complexes. I have learnt that a HADDOCK score is not indicative of binding affinity. I would highly appreciate if if I could receive some guidance on how to go about comparing binding affinities amongst different protein complexes. Thank you.

There is a chaotic nature to the computations. Also since we are using distributed grid infrastructure to run the jobs, each model might come from a different hardware, which does create small differences. Unavoidable unless you run exactly all jobs on the same computer.

As for the binding affinity question, this is a hard problem. We have published several papers on this topic.
Check our PRODIGY papers and also our web server. But realise that the accuracy of such predictors is still quite limited.

Firstly, thank you so much for such a quick response. Since I cannot say anything about binding affinity as such, would a difference in the scores of different complexes at least suggest that the binding affinity is DIFFERENT (not necessarily better or worse)?
While mentioning the scores in my work, I shall definitely explain why there are minor differences in the scores. Thank you.

I will surely check out the papers.

Interesting question. And is answer is probably yes and no.
This has never been tested actually. I am sure you will find case where the score is different and the affinities similar and vice-versa

Rather than using the score, I suggest to use our PRODIGY web server to predict the affinity.
Also do it not for a single model but for multiple models to get an idea of the variation in the prediction.

Just my comment/suggestion on this; let’s say you have one receptor (A) and two ligands (B and C) and AB score is -200 and AC is -100.

Always consider how the score is calculated, the lowest score will tell you the interaction is energetically more favourable, but a more favourable interaction might not correlate directly with the biological process.

Maybe the process is more effective when a given receptor makes fast transient interactions, or maybe when it binds strongly to be translocated to another part in the cell or any things of the sort.

I’d say its important to always analyse any of these metrics in the context of your biological problem. Good luck!

@amjjbonvin and @honoratorv, thank you so much for your responses. They have been of great help, and I will definitely take your inputs into consideration.