Salvatore Benfatto from the Dana-Farber Cancer Institute presented at London Calling 2025 on “Rapid epigenomic classification of acute leukemia.” Benfatto started with a clinical case of a sixty-one year old female patient with intermittent fevers, fatigue, and leukocytosis. Multiple tests had to be run in parallel, including flow cytometry, karyotyping, immunohistochemistry… The process is time and resource intensive. A trained model has been used, and Benfatto wanted to improve the current system. For this, they established a methylation-based acute leukemia reference from public data. DNA methylation was found to resolve disease heterogeneity in acute myeloid leukemia. The next step was to design a machine learning algorithm for nanopore data. The team developed a neural network model for acute leukemia classification. The MARLIN model was forced to perform predictions with masked smaller datasets. MARLIN performance was cross-validated and tested on authentic samples. MARLIN was able to accurately classify acute leukemia from a retrospective cohort study. Next, Benfatto’s team collected samples for a prospective study to determine time necessary for real-time classification. The MARLIN classifier generated high confidence data in only forty minutes of sequencing. The classifier produced results that matched the clinical tests. Benfatto and team want others to use the MARLIN model and try it. The web app can be used to upload bam/methylation results to obtain classifications. Benfatto ended with a photograph of the team next to their sequencer preparing to sequence and use MARLIN on a real patient sample.
