Biotech, AI Company Partner on Covid-19 Alert System

Transmission electron micrograph of a SARS-CoV-2 virus particle, isolated from a patient. (NIAID, NIH)

11 Jan. 2021. A developer of Covid-19 vaccines and an artificial intelligence company are building a system to quickly detect dangerous SARS-CoV-2 variants. A team from biotechnology company BioNTech SE in Mainz, Germany and A.I. developer InstaDeep Ltd in London posted first results from the system on 27 Dec. 2021 in BioRxiv, a pre-print server, while awaiting peer review and formal publication.

BioNTech designed a vaccine technology with messenger RNA that it licensed to global drug maker Pfizer for one of the leading vaccines protecting against Covid-19 disease. The company’s process uses computational techniques to speed development of vaccines, immunotherapies, and diagnostics, including partnerships with other businesses and academic labs. In Nov. 2020, BioNTech began a collaboration with InstaDeep to apply A.I. techniques for drug design, analytics, manufacturing, and supply chain optimization.

InstaDeep develops A.I. solutions for a range of businesses including biotechnology enterprises. The company offers mathematical models for simulating complex protein interactions called DeepChain for designing biologics and diagnostics. In this case, InstaDeep and BioNTech applied A.I. models to quickly assess sequencing data reported on mutations in the SARS-CoV-2 virus responsible for Covid-19 infections.

Trained with thousands of registered mutation sequences

Current methods for evaluating new variants, say the companies, take valuable time that could be saved with a rapid early-warning system, should a dangerous mutation develop. Authors of the paper note that the SARS-CoV-2 virus mutates frequently, with some 12,000 mutations reported in Dec. 2021 alone, yet most mutations are minor and cause no more harm than before. However the need still exists for techniques to quickly identify more dangerous mutations before they begin spreading throughout populations.

The team from InstaDeep and BioNTech first developed a mathematical model of the SARS-CoV-2 spike protein’s structure. They then added transformer language models that modify the protein’s structure with new sequences from mutations. From those models, the researchers assessed SARS-CoV-2 mutations on their ability to evade immune protections and transmissibility/fitness, defined as interactivity between the spike protein and angiotensin-converting enzyme 2 or ACE2 receptors in cells where infections begin. The models were trained with thousands of registered and published mutation sequences. And the team validated the models with antibody neutralization and binding lab tests.

The researchers ran the models on data from World Health Organization reports of mutations since the emergence of SARS-CoV-2. The results show the models identified in minutes 12 of the 13 variants flagged as dangerous by WHO. The authors say an early warning system employing those models would have spotted these mutations an average of two months sooner. That includes the omicron variant, now spreading through the North America, Europe, and elsewhere, which the authors say they detected the same day its sequence became publicly available.

“Early flagging of potential high-risk variants,” says BioNTech CEO Ugur Sahin and the paper’s senior author in a statement, “could be an effective tool to alert researchers, vaccine developers, health authorities, and policy makers, thereby providing more time to respond to new variants of concern.” Karim Beguir, CEO of InstaDeep and first author of the paper adds, “For the first time, high-risk variants could be detected on the spot, potentially saving months of precious time.”

More from Science & Enterprise:

• Essay – Quick, Name the Secretary of HHS
• Foundation Backing Expanded Pandemic Surveillance
• Infographic – Not Ready for Next Pandemic
• CEPI Funds Multi-Variant Covid-19 Vaccine
• Book Review – The Premonition: A Pandemic Story

Disclosure: the author owns shares in Pfizer

We designed Science & Enterprise for busy readers including investors, researchers, entrepreneurs, and students. Except for a narrow cookies and privacy strip for first-time visitors, we have no pop-ups blocking the page, nor distracting animated GIF graphics. If you want to subscribe for daily email alerts, you can do that here, or find the link in the upper left-hand corner of the desktop page. The site is free, with no paywall. But, of course, donations are gratefully accepted.

*     *     *