Fig. 3: Motif deconvolution via ImmuScope on simulated and experimental MA data.
From: Self-iterative multiple-instance learning enables the prediction of CD4+ T cell immunogenic epitopes
a, Schematic of MHC-II motif deconvolution pipeline. b, Performance (AUPR, AUC0.1 and PPV) of the attention-based MIL module across different MHC-II alleles in the simulated data. Box centre line, median; box limits, upper and lower quartiles; whiskers, 1.5× interquartile range. Each data point represents the performance of the corresponding MHC allele (n = 58). c, Heat map of MIL attention weights and actual labels of the simulated data. d, Motif deconvolution logos on Racle__4037_DC heterozygous dataset and KLD analysis of PSFMs from the deconvoluted peptides and MHC-II immunopeptidomics. Panel a created with BioRender.com.
