NCM 2021: Real-time cancer classification with nanopore sequencing

Franz-Josef Müller and Helene Kretzmer presented the potential of ‘real-time cancer classification’ through an engaging video demo of their lab and analysis workflows for brain tumour classification prediction based on a sample’s methylation profile.

Franz-Josef introduced the sequencing set-up of the research workflow, which is optimised ‘for the fastest possible classification of brain tumour types’. Libraries are prepared using the Rapid Sequencing Kit (SQK-RAD004), these are then sequenced on the MinION Mk1B, and Megalodon is used for combined basecalling, alignment, and methylation calling. A set of custom scripts are also applied for tumour classification.

Franz-Josef moved on to the ‘Clinical Demonstrator’ video, called as such because ‘we deeply care about the clinical translation of our basic research and epigenetic features of cancer subtypes’. This began with fresh biopsies taken directly from the operating theatre; these clinical research samples are brought into their sequencing lab for processing via the aforementioned workflow. Franz-Josef suggested that by bringing nanopore sequencing directly to the operating theatre, intraoperative classification could potentially be ‘within reach’. The bioinformatics pipeline is all performed locally; ‘no cloud computers were harmed for this video’. Only 44 minutes after biopsy processing started ‘the first predictions are going to come in’. Franz-Josef stated that these predictions continue to update as more sequence data is generated during the run.

Helene explained that CpGs are filtered within their pipeline to those covered by a commercially available methylation array; in real time, around 500 CpGs are sequenced within the first few minutes of a MinION run, and 2,000–5,000 CpGs within the first 30 minutes. After around 4–5 hours, no new CpGs are sequenced. The training set that was used as the basis of their model for brain tumour classification prediction was derived from Capper et al. (Nature, 2018); this dataset consists of >2,800 samples and covers over 90 different tumour classes.

Helene concluded the presentation by sharing example results, stating that the classification of brain tumour class was stable after only 1 hour of sequencing, and showing how the read out from the pipeline includes a list of tumour classes and their classification probabilities.