Aprendizagem profunda baseada em word embedding associada a técnicas de redução de dimensionalidade aplicada a análise de variantes do SARS-CoV-2

Abstract

This work aims to develop a new proposal to identify and characterize variants associated with the SARS-CoV-2 virus. The proposal uses deep learning based on word embedding associated with unsupervised learning algorithms such as t-Distributed Stochastic Neighbor Embedding (t-SNE) and Principal Component Analysis (PCA). The proposal allows to visualize and characterize the behavior of the variants in the space of two and three dimensions over time. The work presents results from samples of the SARS-CoV-2 virus collected from January to June 2021 and clearly shows the continuous distancing of viral samples due to mutations over time. Thus, the proposal allows the creation of a new analysis tool associated with the emergence of new variants.