Domain-Scan: Combinatorial Sero-Diagnosis of Infectious Diseases Using Machine Learning.
Journal:
Frontiers in immunology
PMID:
33643301
Abstract
The presence of pathogen-specific antibodies in an individual's blood-sample is used as an indication of previous exposure and infection to that specific pathogen (e.g., virus or bacterium). Measurement of the diagnostic antibodies is routinely achieved using solid phase immuno-assays such as ELISA tests and western blots. Here, we describe a sero-diagnostic approach based on phage-display of epitope arrays we term "Domain-Scan". We harness Next-generation sequencing (NGS) to measure the serum binding to dozens of epitopes derived from HIV-1 and HCV simultaneously. The distinction of healthy individuals from those infected with either HIV-1 or HCV, is modeled as a machine-learning classification problem, in which each determinant ("domain") is considered as a feature, and its NGS read-out provides values that correspond to the level of determinant-specific antibodies in the sample. We show that following training of a machine-learning model on labeled examples, we can very accurately classify unlabeled samples and pinpoint the domains that contribute most to the classification. Our experimental/computational Domain-Scan approach is general and can be adapted to other pathogens as long as sufficient training samples are provided.
Authors
Keywords
AIDS Serodiagnosis
Amino Acid Sequence
Antigen-Antibody Reactions
Base Sequence
Communicable Diseases
DNA Barcoding, Taxonomic
DNA, Recombinant
Epitopes
Genetic Vectors
Hepatitis C
Hepatitis C Antibodies
Hepatitis C Antigens
High-Throughput Nucleotide Sequencing
HIV Antibodies
HIV Core Protein p24
HIV Envelope Protein gp160
HIV Infections
Humans
Machine Learning
Oligonucleotides
Peptide Fragments
Peptide Library
Polymerase Chain Reaction
Serologic Tests