Deep learning finds convergent melanocytic morphology despite noisy archival slides

Melanocytic atypia often leads to diagnostic discordance, complicating its prediction by machine learning models. To overcome this, we paired H&E-stained histology images with contiguous or serial sections immunohistochemically (IHC) stained for melanocytic cells via antibodies for MelanA, MelPro, or SOX10. We developed a melanocytic atypia deep learning pipeline from real-world archives of 122 paired whole slide images from 61 confirmed melanoma in situ (MIS) cases at two institutions. Only 37.7% of pairs matched well enough for deep learning; nonetheless, MelanA+MelPro models achieved an average area under the receiver-operating characteristic (AUROC) = 0.948 and area under the precision-recall curve (AUPRC) = 0.611 (9.3× baseline) and SOX10 models achieved AUROC = 0.867 and AUPRC = 0.433 (7.3× baseline). Despite learning from biologically different nuclear versus cytoplasmic IHC stains, convolutional neural network models exhibited a convergent explainable AI rationale. The resulting multi-antibody virtual stains identified cytologic and small-scale architectural features directly from H&E images, supporting pathologists in assessing cutaneous MIS.