Detection of Novel COVID-19 Variants with Zero-Shot Learning

Le résumé fourni par la source

The COVID-19 (COrona VIrus Disease) pandemic has lead to several genes being sequenced, which are then assigned variant names based on their lineages. Several new variants have evolved with time where machine learning algorithms can be used to rapidly assign variants to existing lineages as well as identify new variants. The genetic sequences are in the form of contiguous combinations of characters of the genetic alphabet (A,C,G,T) and the labels of the variants are in the form of Pango lineages. Pango lineages are assigned to each genetic sequence and medical authorities designate specific lineages as Variants of Concern (VoC). Conventional deep learning models that classify COVID-19 variants typically need to learn from several instances of each class to recognize their features. The occurrence of new variants with time necessitates re-training the models with the new data in addition to the previous datasets. Furthermore, it is challenging to name the new variants according to the Pango lineage nomenclature. We propose a zero-shot learning approach where we train on existing COVID-19 variants using a Siamese Neural Network (SNN) and use the information from the learned embeddings in the detection of new COVID-19 variants. We obtained a validation accuracy of 96.42% using our proposed SNN models. We believe that the application of SNNs to the detection of new COVID-19 variants could benefit the existing state-of-the-art deep learning models.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.