Tumors in brain can be classified according to DNA methylation patterns. Accurate molecular classification has implications for patient treatment and can potentially affect surgical strategy. However, access to tumor material is usually not possible before surgery. Methods to classify tumors within a short time-frame (60 – 90 minutes) would allow surgeons to make informed decisions on how to proceed once surgery has started.
Nanopore sequencing is a newly developed technology that allows real-time analysis of DNA and RNA fragments. Low-coverage nanopore sequencing of brain-tumor biopsies and machine learning algorithms have been used to detect DNA methylation patterns and classify CNS tumors. Researchers at the Institute for Surgical Research, Oslo University Hospital have recently shown that the method can be used in the intraoperative setting. Although classification results can currently be returned back to the operating theater in 90 – 120 minutes, this is on the edge of practicality. Any improvement in the speed of return can have a huge impact on how the surgery is performed.
There are considerable gains to be made with regards to the speed of analysis. For example, efforts have been made to leverage graphics processing units (GPUs) in the field of machine learning.
The primary aim of the project is to implement GPU accelerated machine learning algorithms into brain tumor classification based on nanopore sequences. The secondary aim is to set up an automated analytical pipeline to create real-time reports. The candidate will be working with Dr. Einar Vik-Mo (co-supervisor), a neurosurgeon at Oslo University hospital, Dr. Skarphedinn Halldorsson (email) (main supervisor), a molecular biologist at the Institute for Surgical research and prof. Torbjørn Rognes (internal co-supervisor) at the Biomedical Informatics research group (BMI) in the Section for Machine Learning (ML) at the Department of Informatics (IFI).