Thema der Dissertation:
Development of a Naive Bayesian model for tumor type prediction from sparse methylation data Thema der Disputation:
Methylation-Based Brain Tumor Classification: Machine Learning Approaches and Open Challenges
Development of a Naive Bayesian model for tumor type prediction from sparse methylation data Thema der Disputation:
Methylation-Based Brain Tumor Classification: Machine Learning Approaches and Open Challenges
Abstract: DNA methylation is an epigenetic modification whose patterns provide a molecular footprint of tumor identity, reflecting both the cell of origin and alterations acquired during tumorigenesis. Methylation-based classification has therefore become part of routine brain tumor diagnostics. From a machine learning perspective, this poses a challenging supervised learning problem with hundreds of thousands of CpG features, limited reference samples, and many closely related classes. This talk reviews machine learning approaches to this problem and how their design depends on the available data. It starts with established random forest classifiers trained on methylation arrays, which measure a fixed set of CpG sites across samples but take several days to process. Nanopore sequencing allows real-time profiling of DNA methylation, enabling tumor classification during surgery. However, early sequencing data cover only a small, randomly observed subset of CpG sites that is unknown until inference time. This has led to adapted strategies, such as retraining random forests per sample and training neural networks on simulated sparse profiles. Their comparison shows that combining high predictive performance, low latency, computational efficiency, and interpretability within a single strategy remains an open challenge. The second part of the defense presents a probabilistic approach to tumor classification from sparse methylation data, developed in my dissertation.
Time & Location
Nov 06, 2026 | 10:00 AM
Ort: großer Hörsaal
(Fachbereich Mathematik und Informatik, Takustr. 9, 14195 Berlin)
