Thank you very much.
And I will try it immediately.
But now I still want to propose another question. In my case, I found that binding energy and total score were not always consistent with each other. Sometimes, model with lowest binding energy owned quite high total score. So I think maybe there could be a situation that although the complex model with high propensity to reach to the low binding energy model, it was just a transient binding and unstable, concerning the high score(i.e.high total energy). So I think maybe comparing total energy and binding energy together would be better.
Maybe in your case, they had the same trend? How did you treat them?