Ana Flavia is a PhD student at Barretos Cancer Hospital, Brazil, and also a visiting scholar at The University of North Carolina at Chapel Hill. Flavia presented at London Calling 2025 on “Advancing pediatric germ cell tumor classification through nanopore-based transcriptome analysis.” Germ cell tumors (GCT) were described as rare and diagnosed by morphological classification by pathologists. Flavia studied 47 cases of pediatric GCT, either frozen or biobank samples. Some of the samples yielded low RNA or had low tumor volume and were excluded. Flavia used the semi-automated QIASympony for RNA extraction. RNA was quantified by Qubit. cDNA was synthesized and the PCR-cDNA Barcoding kit was used to prepare libraries. MinION flow cells MIN106D were used. Dorado was used as the basecaller with super high accuracy mode. Alignment was performed with Minimap2 and a machine learning model from UNC was used for classification. Most patients were females. N50 was 390 bp for the cryopreserved samples. Samples were grouped by tumor type. Four mixed tumors were identified. The classifier initially had 90.3% accuracy but low prediction probabilities. They then implemented a calibration step. The accuracy was 93.7% with this learning step, and the model had high F1 scores for most subtypes, except embryonal carcinomas. Flavia noted that this was the sample type that only had four. Gene expression patterns were used to train the model only with FFPE samples and other improvements are being made. They have begun using the classifier with external samples for confirmation. This project is highly collaborative with several institutions and countries helping secure samples and improve the classifier.
