COVID-19 coronavirus vaccine design using reverse vaccinology and machine learning

Edison Ong; Mei U Wong; Anthony Huffman; Yongqun He

The Vaxign-ML pipeline computed the protegenicity (protective antigenicity) score and 159 predicted the induction of protective immunity by a vaccine candidate 19 . The training data 160 consisted of viral protective antigens, which were tested to be protective in at least one animal 161 challenge model 30 . The performance of the Vaxign-ML models was evaluated (Table S1 and 162 Figure S1 ), and the best performing model had a weighted F1-score of 0.94. Using the optimized 163 Vaxign-ML model, we predicted three proteins (S protein, nsp3, and nsp8) as vaccine candidates 164 with significant protegenicity scores ( Table 3 ). The S protein was predicted to have the highest 165 protegenicity score, which is consistent with the experimental observations reported in the 166 literature. The nsp3 protein is the second most promising vaccine candidate besides S protein. 167