Error in saliency maps

Dear DeepFRI developers,
There seems to be an error in the saliency maps when the latest release and the downloadable models are used. For example in the case of 2PE5 in the Nat. Comms paper for the DNA-binding GO term (GO:0003677) the DNA-binding domain of the protein is activated (~1-50 residues, Figure 4C), but with the current version of the tool and models, residues that are not involved in DNA-binding are activated:

Best wishes,
George

Reproduce the saliency map for protein 2PE5 and GO term GO:0003677 using the current release and downloadable models, then compare the activated residues with Figure 4C from the Nat. Comms paper. Done means determining why residues outside the DNA-binding region are activated and correcting the discrepancy with a verified reproduction.

Written by the indexing model from the issue text.