BioNTech SE and InstaDeep Ltd. announced the development of a new computational method that analyses worldwide available sequencing data and predicts high-risk variants of SARS-CoV-2. The Early Warning System (EWS) developed in collaboration by BioNTech and InstaDeep is based on artificial intelligence (AI) calculated immune escape and fitness metrics. The new method combines structural modeling of the viral Spike protein and AI algorithms to quickly flag potential high-risk variants entered into SARS-CoV-2 sequence data repositories within less than a day based on metrics scoring their fitness (e.g. ACE2 and variant Spike protein interaction) as well as their immune escape properties. The companies validated these predictions using experimental data generated in-house and publicly available data.

During the trial period, the system has identified >90% of the World Health Organization (WHO)-designated variants (Variants of Concern, VOC; Variants of Interest, VOI; Variants Under Monitoring,VUM) on average two months in advance. WHO-designated variants Alpha, Beta, Gamma, Theta, Eta and Omicron were detected by the EWS in the same week its sequence was first uploaded. The Omicron variant was ranked as a high-risk variant the same day its sequence became available.

The IHU variant observed in France has also been evaluated by the EWS, which highlighted immune escape properties that are relatively similar to Omicron but with significantly lower fitness, making it less of a concern given current data. The results from the study underline that the EWS is capable of evaluating new variants in minutes and risk monitoring variant lineages nearly in real-time. It is also fully scalable as new variant data becomes available.

The Early Warning System (EWS) relies on two approaches: (1) structural modeling of the interaction of the viral Spike protein receptor-binding domain (RBD) with the host cell receptor and scoring the impact of the virus variant in escaping the immune response, and (2) AI-based predictive modeling to extract information from hundreds of thousands of registered virus variants from global sequence repositories. The EWS computes an immune escape score and a fitness (transmissibility potential) prior score. While the immune escape score alone was already highly predictive of the risk, combining these two metrics into a Pareto score provided the best assessment of the risk posed by a given virus variant.

The higher the score, the higher the risk of the variant impacting global health. The EWS approach ranks SARS-CoV-2 variants for immune escape and fitness potential based solely on existing data, and therefore is not dependent on a “wait-and-watch” approach. The EWS was able to distinguish the WHO-designated variants from those that had no designation during a 11-month period, underlining the viable computational model ability to determine variant lineage.

An analysis conducted every week between September 16th, 2020 and November 23rd, 2021 flagged 12 out of 13 WHO-designated variants with an average of 58 days of lead time (two months) before the variants were given their designation. For variants Alpha to Mu, only around 25 cases on average were recorded at the time of them being flagged by the EWS. This is in contrast with the WHO announcements that happened on average when more than 1,500 cases were recorded.

The EWS detected Omicron on the day its sequence was first uploaded as the higher immune escaping variant from over more than 70,000 variants that were discovered between early October 2021 and late November 2021 while also assigning it a high fitness score.