HCV genotyping using models based on deep neural networks with Transformer architecture

• Grant number: 23/14749-5
• Support Opportunities: Scholarships in Brazil - Scientific Initiation
• Effective date (Start): February 01, 2024
• Effective date (End): December 31, 2024
• Field of knowledge: Health Sciences - Collective Health - Public Health
• Principal Investigator: Fernando Martins Antoneli Junior
• Grantee: Ariella Aro
• Host Institution: Escola Paulista de Medicina (EPM). Universidade Federal de São Paulo (UNIFESP). Campus São Paulo. São Paulo , SP, Brazil
• Associated research grant: 20/08943-5 - Investigation of the hosts' induced elements in response to the immunisation with ChAdOx1 nCOV-19 vaccine in a Phase III Clinical Trial, AP.TEM

Abstract

The determination of viral genotypes is an important factor for the response of the patients to the antiviral therapy. Generally speaking, tracking viral infections with simple, rapid, but highly sensitive and specific methods is critical to curb the global burden of antiviral treatments on the healthcare system. Therefore, in addition to determining the viral load, the identification of specific genotypes / subtypes of the virus is essential for the success of possible treatments. Technological advances over the past decade have enabled high-throughput sequencing to produce an abundance of unlabeled, easily available data. To take advantage of this, the bioinformatics field has sought to apply machine learning methods to genomics tasks, often adapting models originally developed in Natural Language Processing for use with genomic data. One of these classes of models is known as 'Transformer Architecture' - for example, the famous ChatGTP (Open AI) is based on this architecture. In this project, we propose to investigate the ability of transformer-based models to produce appropriate representations, trained with unlabeled and labeled data, in order to taxonomically identify viral genotypes / subtypes directly from genomic sequences of arbitrary sizes. To this end, we chose Hepatitis C Virus (HCV) to carry out a 'proof of concept' project on determining viral genotypes / subtypes. We hope that the class of transformer models lends itself well to 'transfer learning', that is, that the obtained representations generalize well to other types of viruses, especially RNA viruses.