A deep learning approach predicting the activity of COVID-19 therapeutics and vaccines against emerging variants

Abstract

Abstract Understanding which viral variants evade neutralization is crucial for improving antibody-based treatments, especially with rapidly evolving viruses like SARS-CoV-2. Yet, conventional assays are labor intensive and cannot capture the full spectrum of variants. We present a deep learning approach to predict changes in neutralizing antibody activity of COVID-19 therapeutics and vaccine-elicited sera/plasma against emerging viral variants. Our approach leverages data of 67,885 unique SARS-CoV-2 Spike sequences and 7,069 in vitro assays. The resulting model accurately predicted fold changes in neutralizing activity (R2 = 0.77) for a test set (N = 980) of data collected up to eight months after the training data. Next, the model was used to predict changes in activity of current therapeutic and vaccine-induced antibodies against emerging SARS-CoV-2 lineages. Consistent with other work, we found significantly reduced activity against newer XBB descendants, notably EG.5, FL.1.5.1, and XBB.1.16; primarily attributed to the F456L spike mutation.