Extended Data Fig. 2: Ablation analysis of ImmuScope-EL modules.
From: Self-iterative multiple-instance learning enables the prediction of CD4+ T cell immunogenic epitopes
a, AUC0.1 performance of SA, SA + MA-MIL, and ImmuScope-EL. The P values were calculated by the two-sided Wilcoxon signed rank test (n = 58): SA vs. SA + MA-MIL (\(P=1.3\times {10}^{-10}\)) and SA + MA-MIL vs. ImmuScope-EL (\(P=1.7\times {10}^{-8}\)). Box center line, median; box limits, upper and lower quartiles; whiskers, 1.5×interquartile range; ****P < 0.0001. Each data point represents the performance of the corresponding MHC allele. b, AUPR performance of SA, SA + MA-MIL, and ImmuScope-EL, calculated per MHC-II allele, only including alleles with at least 25 positive samples and a minimum of 30 total samples. c, AUC0.1 performance of the ImmuScope-EL model for different MHC alleles. The bars represent the mean performance values by 1,000 bootstrap iterations, and the error bars indicate the 95% CIs. Each data point represents the performance of the corresponding MHC allele. d, UMAP visualization of instance embeddings from the SA and ImmuScope-EL models on the BoLA-DRB3 subset.
