We are leading the Swiss Personalized Oncology (SPO) consortium that assembles medical oncologists specialized in precision oncology from the Swiss university hospitals together with high profile scientists involved in specific –omics analyses (single-cell proteomics, single-cell transcriptomics, drug screening, digital pathology). SPO aims to bring -omics results back to patient care by ensuring national data interoperability, rapid turnaround times, and automated generation of clinical reports that will support therapeutic decision making within the existing molecular tumor boards. In addition, SPO aims to understand the factors associated with primary or acquired resistance to immunotherapy in order to identify predictive biomarkers to help treatment decisions.

We develop interactive data visualization tools to make data available in a convenient form to clinicians and researchers. Our tools allow quickly assembling virtual cohorts of real-world patients using searches in structured or free-text fields of the electronic patient records, and to curate data on treatments, radiological response, toxicities, comorbidities and more. We then apply cutting edge supervised and unsupervised methods to select covariates of interest and to derive predictors of patient outcomes.

The analysis of histological sections of tumor tissue is widely used as a diagnostic tool for various forms of cancers. In collaboration with the Pathologie department, we aim to exploit this rich source of patient-specific information much further by combining it with modern image processing and machine learning techniques. In particular, multiplexed immunohistochemistry (mIHC) images can be analyzed in terms of complex features, such as density profiles, cell type co-localization, network connectedness, which can be subsequently fed to machine learning algorithms.

While current outcome assessments from CT or PET-CT imaging rely on aggregate grading such as RECIST, semi-automated image analysis techniques make it possible to follow the evolution of single lesions. In addition to their size, a large number of so-called radiomics features can be calculated, which capture shape, intensity distributions, texture measures, etc… Such radiomics features can then be used, in collaboration with experts in Radiology and Nuclear Medicine, to inform models of tumor dynamics, or to train predictors of response to treatment.