I tried correlating two bigwig tracks (fold change over control and signal p-value) around the ATAC-seq peaks from the GM12878 ATAC-seq ENCODE experiment and similarly for K562 ATAC-seq ENCODE experiment. However, the rank correlation in both these cases is only around 0.8. Could anyone help me understand why this might be the case, especially since both tracks are created from the same bam files and hence, the same control?
There are two degrees of freedom for the p-value: the magnitude of the difference between signal and background, and the variability of the signal and the background. Cases where the same fold change have different p-values are cases where there are differences in variability. The fold-change is providing you information about magnitude, and the p-value is providing you information about confidence.