Figure 4. Application of the fine-tuned language model-based classifiers to vaccine response repertoire datasets.

(A) Mean predicted probability of SARS-CoV-2 S protein binders by fine-tuned language model S protein classifier of the receptors from peripheral blood samples 28 days after SARS-CoV-2 vaccination (Day 28), compared with the pre-pandemic repertoire datasets (Control). Samples from the same donor were connected by lines. Paired Wilcoxon rank sum test was used to obtain the significance level of the increase in mean predicted probability at day 7 (ns: p > 0.05, *: p <= 0.05, **: p <= 0.01, ***: p <= 1e-3, ****: p <= 1e-4). (B) Mean predicted probability of S protein binders applied to lymph node repertoires 28, 35, 60, 110 days after SARS-CoV-2 vaccination. (C) Mean predicted probability of influenza HA binders by language model fine-tuned on HA classification task applied to peripheral blood repertoires before and seven days after influenza vaccination.