Pillar 1 - Machine learning and bioinformatics for omics data –In our lab, we are leveraging the latest AI developments for personalized medicine. For cancer patient, for instance, individualized decisions can be based on a thorough characterization of the tumour at a molecular level. However, recent artificial intelligence (AI) models, that work so well on images and text, struggle in dealing with the complexity of these molecular data. When it comes to patient-derived molecular profiles there is simply not enough patient data to use modern AI models. To address this, we are trying a new approach. Instead of immediately training on patient data, we first want to teach an AI about molecular disease biology by creating so-called foundation models based on massive amounts of biomolecular data from single cells, molecular profiles of healthy and sick tissues and biomolecular network information. Based on these data we use self-supervised learning (SSL), an important driver of AI. Once the AI has some ‘common sense’ about molecular disease biology we can finally train it to make predictions such as “what drug should this patient receive?” or “what tumour subtype does this patient have?”. Furthermore, we are studying how mutations and modifications affect the functions encoded in the genome and contribute to disease. For instance, we pioneered deep learning modelling of Massively Parallel Reporter Assay (MPRA) data to predict promoter or enhancer activity. These semi-supervised AI models can be used for predicting the effect of non-coding mutations in cancer based on DNA sequence alone. With this research line we aim to leverage the enormous progress in AI and thereby bring truly personalized medicine one step closer.   

Pillar 2 - Bringing data-driven cancer diagnostics solutions to the clinic – Our lab aims to short-cut the route from fundamental research results to patient benefit and actively contributes to clinical application of the research. My lab has embraced nanopore sequencing to lower the threshold for routine diagnostic sequencing and developed a nanopore based liquid biopsy test (Published in NPJ Genomic Medicine in 2021), filed a patent on the underlying technology (US20210180109A1) and founded a start-up company Cyclomics BV, aimed at developing the next generation of liquid biopsies. Most recently, we have submitted our work demonstrating that nanopore sequencing can be used for ultra-rapid methylation-based brain cancer classification during the resection surgery and that, by using our deep learning model “Sturgeon”, turnaround times of less than 90 minutes are possible (Published in Nature in 2023).