PhD candidate ‘AI-based pan-cancer prediction of 'omics biomarkers from histopathology’

Molecular alterations, such as Microsatellite Instability (MSI), Homologous Recombination Deficiency (HRD), and specific gene mutations, play a critical role in modern oncology. They are essential 'omics biomarkers for selecting patients for targeted therapies and immunotherapies. Currently, determining the presence of these biomarkers requires expensive, tissue-consuming, and time-intensive molecular testing, such as next-generation sequencing. However, these underlying genomic alterations often manifest as distinct morphological patterns within the tumor and its microenvironment.

In this project, you will harness the power of artificial intelligence to predict these crucial 'omics biomarkers directly from routinely available hematoxylin and eosin (H&E)-stained histopathology slides. Because biomarkers like MSI have therapeutic relevance across many solid tumor types, your research will take a pan-cancer approach. You will leverage state-of-the-art pathology foundation models and weakly-supervised learning techniques to extract robust representations from gigapixel whole-slide images, learning directly from clinical ground-truth labels without the need for exhaustive manual annotations.

Furthermore, clinical adoption of AI requires transparency. A major focus of your project will be developing and evaluating explainability methods. By "opening the black box," you will help pathologists and oncologists understand the morphological features driving the AI’s predictions, building clinical trust and potentially discovering novel visual correlates of underlying genomic alterations.

Tasks and responsibilities

• Develop and optimize weakly-supervised deep learning algorithms and adapt pathology foundation models to predict 'omics biomarkers (e.g., MSI) across diverse cancer types.
• Investigate and pioneer explainability techniques to make complex neural network predictions transparent and interpretable for clinical end-users.
• Process and analyze massive, multi-centric datasets of oncological whole-slide images with matched molecular ground-truth data.
• Collaborate closely with pathologists, oncologists, and (inter)national machine learning researchers to validate the clinical and biological relevance of your developed algorithms.
• Have fun interactions with colleagues, present at local and (inter)national conferences, publish in high-impact technical and medical journals, and develop yourself as an independent researcher.