CVSimP: An approach for predicting proteins’ structural similarity using one-shot learning

Proteomics is considered an important field from the computational biology. It deals with answering some challenging questions regarding proteins structure. On the other hand, it is nowadays well known that computer vision and machine learning can be used together to build powerful systems for analysing complex image data. This paper introduces a new approach for grouping proteins based on their internal structure similarity. We propose a system based on one-shot learning and siamese convolutional neural networks for dealing with this task. The system analyses graphical representations of proteins obtained from the Protein data bank for clustering them together based on their structural similarities. The experimental results highlight that CVSimP outperforms, in terms of F-measure, similar related work from the literature.