GNN-based Antibody Structure Prediction using Quaternion and Euler Angle Combined Representation

In recent years, antibodies have gained much attention because they are immune system proteins playing essential roles in vaccine development. Since the 3D structures of the proteins are related to various biological interactions, the importance of 3D protein structures has been highlighted from various researches and proper representation of orientations has become a significant issue in antibody structure prediction tasks. In this paper, we use the combination of Quaternion and Euler angle methods to obtain a rich representation of the rotation between orientations. The main advantage of the method is that both Quaternion and Euler angle methods can compensate for each other to achieve higher interpretability and expressivity than the single use each method. In the experiment, we compare the results of the proposed method to the competing method, and the experimental results show further improvement in Root Mean Square Deviation (RMSD) up to 4.92%, demonstrating the effectiveness and robustness of our proposed method.