Machine Learning-based Analysis of Sequenced Antibodies to Determine Broadly Neutralizing anti-DENV Antibodies

Abstract

Background: The dengue virus is a global threat to public health as roughly half of the world's population is living in a risk area and as of today, there is no vaccine for disease prevention against all dengue serotypes (WHO, 2019). Advancements in high throughput sequencing enable efficient analysis of human immune repertoire data to obtain more insights about the immune disease. As the complexity of sequencing data is very high, artificial intelligence-based analysis could help overcome this complexity and ultimately lead to the identification of broadly neutralizing antibodies against the dengue virus. Methods: Various supervised machine learning (ML) methods were trained with publicly available sequencing datasets of serotype-specific, dengue-challenged individuals. Additionally, a control group of sequences from non-immunized healthy individuals was used (repertoire-level classification, Godoy-Lozano et al., 2016; Huang et al., 2017; Parameswaranet al., 2013). Besides considering different machine learning architectures, various encoding methods were taken into consideration and a novel physicochemical properties-based encoding method was introduced. Subsequently, similarity networks were used to justify the predictions before the models were trained to classify whether single antibodies could potentially bind to dengue antigens or not (sequence-level classification)....