Erasmus MC Cancer Institute

Sturgeon V2, a classifier for the sparse methylation profiles produced during intraoperative sequencing of CNS tumors, is trained on nearly 10,000 relabeled profiles instead of the 2,801-sample 2018 reference set that all publicly available sparse-profile classifiers had relied on. In a retrospective cohort of 94 pediatric intraoperative cases it was confidently correct in 85, versus 78 for Sturgeon V1, and in 1,880 simulated sequencing runs it cut the error rate among confident calls from 7.1% to 2.8%.

Sturgeon V2, a classifier for the sparse methylation profiles produced during intraoperative sequencing of CNS tumors, is trained on nearly 10,000 relabeled profiles instead of the 2,801-sample 2018 reference set that all publicly available sparse-profile classifiers had relied on. In a retrospective cohort of 94 pediatric intraoperative cases it was confidently correct in 85, versus 78 for Sturgeon V1, and in 1,880 simulated sequencing runs it cut the error rate among confident calls from 7.1% to 2.8%.