Dongsheng Yuan

Background: Although cavitating ultrasonic aspirators are commonly used in neurosurgical procedures, the suitability of ultrasonic aspirator-derived tumor material for diagnostic procedures is still controversial. Here, we explore the feasibility of using ultrasonic aspirator-resected tumor tissue to classify otherwise discarded sample material by fast DNA methylation-based analysis using low pass nanopore whole genome sequencing. Methods: Ultrasonic aspirator-derived specimens from pediatric patients undergoing brain tumor resection were subjected to low-pass nanopore whole genome sequencing. DNA methylation-based classification using a neural network classifier and copy number variation analysis were performed. Tumor purity was estimated from copy number profiles. Results were compared to microarray (EPIC)-based routine neuropathological histomorphological and molecular evaluation. Results: 19 samples with confirmed neuropathological diagnosis were evaluated. All samples were successfully sequenced and passed quality control for further analysis. DNA and sequencing characteristics from ultrasonic aspirator-derived specimens were comparable to routinely processed tumor tissue. Classification of both methods was concordant regarding methylation class in 17/19 (89%) cases. Application of a platform-specific threshold for nanopore-based classification ensured a specificity of 100%, whereas sensitivity was 79%. Copy number variation profiles were generated for all cases and matched EPIC results in 18/19 (95%) samples, even allowing the identification of diagnostically or therapeutically relevant genomic alterations. Conclusion: Methylation-based classification of pediatric CNS tumors based on ultrasonic aspirator-reduced and otherwise discarded tissue is feasible using time- and cost-efficient nanopore sequencing.

DNA methylation-based classification of brain tumors has emerged as a powerful and indispensable diagnostic technique. Initial implementations have used methylation microarrays for data generation, but different sequencing approaches are increasingly used. Most current classifiers, however, rely on a fixed methylation feature space, rendering them incompatible with other platforms, especially different flavors of DNA sequencing. Here, we describe crossNN, a neural network-based machine learning framework which can accurately classify tumor entities using DNA methylation profiles obtained from different platforms and with different epigenome coverage and sequencing depth. It outperforms other deep- and shallow machine learning models with respect to precision as well as simplicity and computational requirements while still being fully explainable. Validation in a large cohort of >1,900 tumors profiled using different microarray and sequencing platforms, including low-pass nanopore and targeted bisulfite sequencing, demonstrates the robustness and scalability of the model.