According to the zero_shot_mutation.ipynb, the model outputs multiple feature values, which are used to compute the correlation with ΔΔG. However, in Fig. 5 of the article, there is only a single overall score that is compared with the ΔΔG difference. How is this overall score specifically obtained?