Abstract
Computational prediction of T cell epitopes is a crucial component in the development of novel vaccines. T cells in a healthy vertebrate host can recognize as non-self only those peptides that are present in the parasite's proteins but absent in the host's proteins. This principle enables us to determine the current and past host specificity of a parasite and to predict peptides capable of eliciting a T cell response. Building upon the detailed mapping of T cell clone specificity for Severe Acute Respiratory Syndrome Coronavirus 2 (SARS-CoV-2) antigens, we employed Monte Carlo tests to determine that empirically confirmed T cell-stimulating peptides have a significantly increased proportion of pentapeptides, hexapeptides and heptapeptides not found in the human proteome (P < 0.0001, Cohen's d > 4.9). We observed a lower density of potential pentapeptide targets for T cell recognition in the spike protein from the human-adapted SARS-CoV-2 ancestor compared to 10 other SARS-CoV-2 proteins originating from the horseshoe bat-adapted ancestor. Our novel method for predicting T cell immunogenicity of SARS-CoV-2 peptides is four times more effective than previous approaches. We recommend utilizing our theory-based method where efficient empirically based algorithms are unavailable, such as in the development of certain veterinary vaccines, and combining it with empirical methods in other cases for optimal results.