Summary: Researchers at USC report an AI-driven system that can rapidly analyze potential SARS-CoV-2 variants and compress vaccine design cycles from months to minutes, enabling faster identification of promising vaccine candidates.
Source: USC
USC engineers have created an artificial intelligence framework that accelerates vaccine design and helps counter emerging coronavirus mutations.
A research team at the USC Viterbi School of Engineering applied machine learning to streamline the vaccine discovery workflow for SARS-CoV-2. Their approach shortens the time needed to evaluate and prioritize vaccine subunits, enabling rapid selection of the most promising candidates for laboratory validation and eventual clinical trials.
The system, described in Scientific Reports, rapidly filters large numbers of potential vaccine components to identify those most likely to generate a protective immune response. According to the study, the AI model can perform computational vaccine-design cycles in seconds or minutes—work that historically took months or years—giving scientists a practical tool to keep pace with an evolving virus.
“This AI framework, applied to the specifics of this virus, can provide vaccine candidates within seconds and move them to clinical trials quickly to achieve preventive medical therapies without compromising safety,” said Paul Bogdan, associate professor of electrical and computer engineering at USC Viterbi and corresponding author of the study. “Moreover, this can be adapted to help us stay ahead of the coronavirus as it mutates around the world.”
When the researchers applied the model to the SARS-CoV-2 spike protein, the AI eliminated roughly 95% of theoretical compounds that would be unlikely to work, rapidly narrowing the field. The method produced 26 top-ranking vaccine subunit candidates; from these, the team selected 11 high-quality components to build a multi-epitope vaccine aimed at disrupting the spike protein regions the virus uses to bind and enter human cells.
Multi-epitope vaccines combine several B-cell and T-cell epitopes—the specific parts of a pathogen recognized by the immune system—to produce broader and potentially more durable immune protection. The USC approach evaluated linear B-cell epitopes as well as cytotoxic T lymphocyte (CTL) and helper T lymphocyte (HTL) epitopes to assemble a balanced design while checking for antigenicity, allergenicity, toxicity, and other important properties.
A key advantage of this AI-assisted pipeline is speed. The team reports that constructing a new multi-epitope vaccine for a novel virus variant can take less than a minute computationally, with quality validation completed within an hour. That contrasts sharply with traditional vaccine development methods that require culturing and inactivating virus—processes that can take a year or longer before candidate vaccines reach clinical testing.
USC method could help counter COVID-19 mutations
Rapid adaptability is crucial as SARS-CoV-2 continues to mutate worldwide. Variants first identified in the United Kingdom, South Africa and Brazil have shown increased transmissibility, prompting concerns that some existing vaccines may be less effective against certain mutations. The USC framework is designed to be updated quickly with new genomic information so it can propose revised vaccine combinations that address emerging changes in the virus.
In their study, the authors used a relatively small set of epitopes—one B-cell and one T-cell epitope in an illustrative example—but note that expanding the dataset and testing many more combinations would enable a more comprehensive vaccine-design tool. The framework is scalable and, according to the paper, capable of accurate prediction across datasets that include hundreds of thousands of proteins.
“The proposed vaccine design framework can tackle the three most frequently observed mutations and be extended to deal with other potentially unknown mutations,” Bogdan said, emphasizing the model’s flexibility for ongoing surveillance and rapid response.
The computational findings rely on publicly available immunological repositories. Key inputs included the Immune Epitope Database (IEDB), which compiles experimentally validated epitopes from thousands of species, and viral sequence resources such as the National Center for Biotechnology Information for SARS-CoV-2 genome and spike protein data.
Worldwide, COVID-19 has resulted in tens of millions of confirmed cases and significant mortality, along with major social and economic disruption. Rapid, adaptive vaccine design tools can help public health responses by shortening the time from variant detection to candidate vaccine selection.
The study was authored by Paul Bogdan, Zikun Yang and Shahin Nazarian of the Ming Hsieh Department of Electrical and Computer Engineering at USC Viterbi.
Funding: Support for the research came from the National Science Foundation (NSF) under the Career Award (CPS/CNS-1453860) and NSF grants (CCF-1837131, MCB-1936775 and CNS-1932620); a U.S. Army Research Office grant (W911NF-17-1-0076); a Defense Advanced Research Projects Agency (DARPA) Young Faculty Award and Director Award grant (N66001-17-1-4044); and a Northrop Grumman grant.
About this artificial intelligence and coronavirus research news
Source: USC
Contact: Gary Polakovic – USC
Image: The image is in the public domain
Original Research: Open access. “An in silico deep learning approach to multi-epitope vaccine design: a SARS-CoV-2 case study” by Zikun Yang, Paul Bogdan & Shahin Nazarian. Scientific Reports
Abstract
An in silico deep learning approach to multi-epitope vaccine design: a SARS-CoV-2 case study
SARS-CoV-2 has spread rapidly across the globe, causing widespread illness and significant loss of life. In the absence of universally effective treatments, vaccines remain the primary tool to curb transmission. This study presents DeepVacPred, an in silico framework that combines immunoinformatics and deep neural networks to predict and design multi-epitope vaccines from the SARS-CoV-2 spike protein sequence. DeepVacPred identified 26 candidate subunits and, after extended computational evaluation of B-cell, CTL, and HTL epitopes, selected 11 high-quality components to assemble a multi-epitope vaccine. The design was evaluated for population coverage, antigenicity, allergenicity, toxicity, physicochemical characteristics and secondary structure using state-of-the-art bioinformatics tools. The predicted 3D structure was refined and validated in silico, and codon optimization was applied to support cloning and expression. The resulting 694-amino-acid multi-epitope vaccine construct contains numerous B-cell, CTL and HTL epitopes and is positioned for further laboratory and clinical evaluation. The framework also incorporates RNA mutation analysis to ensure the designed vaccine can address recently observed viral mutations.