![]() | Only 14 pages are availabe for public view |
Abstract Assembly of capsid virus is a crucial step in virus life cycle. Without this step, virus would not replicate itself to hijack other cells and its life cycle would end. Many researchers studied virus structural shape and its dynamics to understand the behavior of virus. A small virus capsid contains identical asymmetric units that are packed in regular manner. Every icosahedral virus has two types of symmetry, regular symmetry and noncrystallographic symmetry. One asymmetric unit and some rotation matrices are needed to form the whole capsid. These rotation matrices define the location of adjacent asymmetric unit.This thesis focuses on the structural shape of Icosahedral viruses and prediction of symmetries in their capsids.Two approaches are carried out to study the construction of a crystal asymmetric unit. The first approach predicts the full rotation matrix (4x4 matrix). The second approach predicts the rotation angles and translation vector. In each approach, we are applying convolution neural network (CNN) and fully connected neural network (FNN). Spatial geometry and biological characteristics were collected for each icosahedral capsid virus from the Protein Data Bank (PDB). Using visualization technique, the results were promising; as in the approach that predicts angles, FNN model accuracy reached 89% and reached 84% in CNN model in the same approach. While the second approach had a lower accuracy percentage; as it reached 67% in FNN model and 45% in CNN model. FNN models in general gave better performance in accuracy and 0.25% less in time and 0.90% less memory consumption than CNN models |