Back

TUCAN: Ultra-fast methylation-based classification of pediatric solid tumors and lymphomas

Jongmans, M.; van Tuil, M.; de Ruijter, E.; Hiemcke-Jiwa, L.; Flucke, U.; de Krijger, R.; Scheijde-Vermeulen, M.; Kusters, P.; van Ewijk, R.; Merks, H.; van Noesel, M.; Pages-Gallego, M.; Vermeulen, C.; Tops, B.; de Ridder, J.; Kester, L.

2026-03-26 oncology
10.64898/2026.03.24.26348466 medRxiv
Show abstract

The high heterogeneity of pediatric cancers presents significant diagnostic challenges, underscoring the need for accurate classification. Although molecular profiling supports first-line diagnostics and guides treatment, it can delay final diagnosis. While Nanopore-based methylation analysis has enabled rapid CNS tumor diagnosis, its application to pediatric solid tumors and lymphomas has remained largely unexplored. We developed Tucan, a deep-learning classifier trained on 3,818 methylation array profiles representing 84 subtypes, designed to classify tumors from sparse Nanopore methylation data. In retrospective validation (n=514), Tucan generated confident predictions (CFT[≥] 0.7) within 30 minutes of sequencing in 385 cases, achieving 372 correct diagnoses (F1-score: 0.98). In prospective testing (n=74; 63 classifiable), 52 samples reached the confidence threshold with 96% accuracy, confirming the original diagnosis in 47 cases and correctly refining or revising it in three. Together, Tucan enables rapid, high-confidence molecular classification of pediatric solid tumors and lymphomas.

Matching journals

The top 9 journals account for 50% of the predicted probability mass.

50% of probability mass above

"Similar papers" are the closest papers from that journal in the model's embedding space. They show what the match is built on, but the ranking comes mostly from a classifier over the whole training set, not from these examples alone.