Patterns of Volatility Across the Spike Protein Accurately Predict the Emergence of Mutations within SARS-CoV-2 Lineages.

New lineages of SARS-CoV-2 are constantly emerging. They contain mutations in the spike glycoprotein that can affect virus infectivity, transmissibility, or sensitivity to vaccine-elicited antibodies. Here we show that the emergence of new spike variants is accurately predicted by patterns of amino acid variability (volatility) in small virus clusters that phylogenetically-precede or chronologically-predate such events. For each spike position, volatility within the virus clusters, volatility at adjacent positions on the three-dimensional structure of the protein, and volatility across the network of co-volatile sites describe its likelihood for mutations. By combining these variables, early-pandemic sequences accurately forecasted mutations in lineages that appeared 6-13 months later. The patterns of mutations in variants Alpha and Delta, as well as the recently-appearing variant Omicron were also predicted remarkably well. Importantly, probabilities assigned to spike positions for within-lineage mutations were lineage-specific, and accurately forecasted the observed changes. Sufficient antecedent warning of the imminent changes in SARS-CoV-2 lineages will allow design of immunogens that address their specific antigenic properties. SIGNIFICANCE: New variants of SARS-CoV-2 continue to emerge in the population. Due to mutations in the spike protein, some variants exhibit partial resistance to therapeutics and to the immunity provided by COVID-19 vaccines. Thus, there is a need for accurate tools to forecast the appearance of new virus forms in the population. Here we show that patterns of amino acid variability across the spike protein accurately predict the mutational patterns that appeared within SARS-CoV-2 lineages with considerable advance warning time. Interestingly, mutation probabilities varied greatly between lineages, most notably for critical sites in the receptor-binding domain of spike. The high predictive capacity of the model allows design of vaccines that address the properties of variants expected to emerge in the future.