Enformer gene expression score prediction

Hello,

I am learning Enformer and interestong in your paper "Effective gene expression prediction from sequence by integrating long-range interactions". It is absolutely a wonderful job. I am wondering why do you measure the performance of model based on the pearson correlation of predicted and real peak value (Figure 1B in the article) ? I think pearson correlation is not an intuitive indicator, it can not measure the absolute difference of predicted and real peak values.