Proteins are essential components of our lives. This has made their classification a very active research field. Most research focuses on classifying proteins based on their structure, overlooking the surface, which is a challenging aspect of research due to its complexity. However, by relying only on the structural aspect, relevant information will be lost. The main objective of this paper is to propose a new machine learning method for protein classification using Graph Convolutional Neural Networks (GCNNs) applied to 3D files. The use of GCNNs will enable us to efficiently combine these informations with structure information extracted from Protein Data Bank (PDB) files to improve protein representation and enhance accuracy, and then train a classifier to organize and classify the proteins. Additionally, we devide our system into three different modules to ensure the interpretation of the classification results, and ensuring that it works even when one type of information is missing. Our method achieves an average accuracy of 94% on the SHREC19 dataset, and also demonstrates efficient execution times. Finally, we employed a feature importance technique to highlight the most relevant information for deeper insight.