Proteins form the basis of all biochemical processes in an organism. Understanding the structure of a protein is integral to understanding the proteins function. The recent introduction of AlphaFold has lead to a huge performance jump in computational protein structure prediction. However, there can still be errors in the predicted structures, even if it looks correct at a first glance. This work focuses on the protein structure of Cryptochrome 4 (Cry4), which has been proposed to function as a magnetosensory protein in birds. The AlphaFold predicted structure of pigeon and European robin Cry4 shows a substantial difference in structural stability in an area supposedly relevant for magnetoreception. This work suggests an addition to AlphaFold in form of an additional loss term based on the CHARMM36 forcefield, which provides an estimation of the energies acting on the atoms of the protein. This novel forcefield loss performs as well as the integrated structural loss term in AlphaFold and, given more computational power, might be able to provide a different focus for AlphaFold in general, leading to a higher accuracy in the structural flexibility of AlphaFold predictions.