Exploring epitope and functional diversity of anti-SARS-CoV2 antibodies using AI-based methods

Abstract

SummarySince the beginning of the COVID19 pandemics, an unprecedented research effort has been conducted to analyze the antibody responses in patients, and many trials based on passive immunotherapy — notably monoclonal antibodies — are ongoing. Twenty-one antibodies have entered clinical trials, 6 having reached phase 2/3, phase 3 or having received emergency authorization. These represent only the tip of the iceberg, since many more antibodies have been discovered and represent opportunities either for diagnosis purposes or as drug candidates. The main problem facing laboratories willing to develop such antibodies is the huge task of analyzing them and choosing the best candidate for exhaustive experimental validation. In this work we show how artificial intelligence-based methods can help in analyzing large sets of antibodies in order to determine in a few hours the best candidates in few hours. The MAbCluster method, which only requires knowledge of the amino acid sequences of the antibodies, allows to group the antibodies having the same epitope, considering only their amino acid sequences and their 3D structures (actual or predicted), and to infer some of their functional properties. We then use MAbTope to predict the epitopes for all antibodies for which they are not already known. This allows an exhaustive comparison of the available epitopes, but also gives a synthetic view of the possible combinations. Finally, we show how these results can be used to predict which antibodies might be affected by the different mutations arising in the circulating strains of the virus, such as the N501Y mutation that has started to spread in Great-Britain.

Integrating artificial intelligence-based epitope prediction in a SARS-CoV-2 antibody discovery pipeline: caution is warranted

“…Unique to our approach is that we built in an in-silico step, early on in the discovery pipeline, in an attempt to be able to select promising antibody candidates, binding different regions in the SARS-CoV-2 RBD, without the need for time-consuming experimental analyses. Using a combination of algorithms, the commercialized MAbSilico artificial intelligence (AI)-based method that we used 64 allows quick similarity analysis of large antibody sequence datasets, via the enumeration of common subsequences in the CDRs and without the need for structural data. Via this in-silico approach, a similarity matrix between the 398 antibodies was generated ( Fig.…”

“…Epitope mapping using artificial intelligence-based methods were not performed in-house but outsourced to MAbSilico, a company that applies a range of self-developed algorithms to establish computational epitope binning and mapping as described by Dumet, Jullian, 30 amongst others.…”

Severe COVID-19: Drugs and Clinical Trials

“…Dumet et al carried out an in silico analysis and described the possible mAbs combinations that might represent new therapeutic opportunities, discarding those that may be ineffective due to the possible overlapping of epitopes. An example of mAbs that do not overlap the same epitope and are currently being tested as a cocktail in a phase 3 clinical trial is tixagevimab + cilgavimab, while tevesimab + bamlanivimab and casirivimab + imdevimab have already been authorized [83].…”

Arsenal of Nanobodies for Broad-Spectrum Countermeasures against Current and Future SARS-CoV-2 Variants of Concerns

“…The binding profiles for the strongest neutralizers in bins 2/3/4 overlap the ACE2 binding site 10 and known conformational hotspots 60 . It was not possible to further resolve epitope diversity within the context of this dataset, however it is evident that a range of binding patterns exists 61,62 , and there is a correlation between stabilizations spanning mutations in VoCs and the loss/attenuation of neutralization (Fig. 5B).…”