deepLIFT scores always central to genomic sequence

[PelFritz]

Hello, I have recently observed something strange when applying deeplift to interpret my models trained to differentiate bound from unbound regions of the genomes using binding profiles similar to ATAC-seq (the profile gets the binding of all proteins in the genome). When i ran deeplift it always created scores that on average were at the center of the sequence. Even when i obtain sequences from the genome were the true bound area was shifted to the left or right, I still get scores that are centered.

Any ideas as to this strange behaviour, will be helpful. Thanks Best Fritz