Back to overview

Trans-omic approach to colorectal cancer: an integrative computational and clinical perspective

Applicant Zlobec Inti
Number 193832
Funding scheme Sinergia
Research institution Institut für Pathologie Universität Bern
Institution of higher education University of Berne - BE
Main discipline Interdisciplinary
Start/End 01.10.2020 - 30.09.2024
Approved amount 2'875'765.00
Show all

All Disciplines (5)

Clinical Cancer Research
Molecular Biology
Information Technology

Keywords (7)

colorectal cancer; histological images; genomics; computation; trans-omics; ex-vivo drug screening; artificial intelligence

Lay Summary (German)

Dickdarmkrebs (DK) ist eine wichtige Ursache der krebsbedingten Sterblichkeit in der Schweiz. Mit den Consensus Molecular Subtypes (CMS) wurde die erste umfassende molekulare Klassifikation mit klinischen Implikationen erstellt, dennoch fehlen einige Aspekte. Wir verwenden einen transomischen Ansatz zur Verbesserung der Stratifizierung, Prognose und Behandlungsvorhersage von DK-Patienten.
Lay summary
Wir untersuchen Varianten von Single Tandem Repeats (STR), die der Tumorentstehung einen selektiven Vorteil verschaffen. Wir verwenden einen neuen evolutionären Modellansatz um die DK "Risiko"-Varianten an klinischen Proben zu validieren. Für unsere histiomische Strategie, wenden wir Methoden des «Deep Learnings» auf histologischen Bildern an, um CMS-Gruppen vorherzusagen. Diese werden auf verschiedene klinische Szenarien angewendet. Die klinische Relevanz der intra- und inter-tumoralen Heterogenität wird evaluiert. Wir nutzen die Pharmakogenomik, um eine experimentelle Plattform für das bildbasierte Ex-vivo-Drogenscreening von Patientenbiopsien bei Lebermetastasen von DK-Patienten zu etablieren. Ex-vivo-Arzneimittelreaktionen werden zur Vorhersage des klinischen Ansprechens eingesetzt, um molekulare Einblicke in DK-Zellen je nach CMS-Klassierung zu erhalten. Zur Vorhersage der CMS-Klassifikation und anderer klinischer Variablen werden alle Daten in einen von künstlicher Intelligenz gesteuerten multimodalen Algorithmen integriert und trainiert. Wir werden Interpretierbarkeitstechniken auf «Deep Learning» Algorithmen untersuchen, um molekularen und morphologischen Muster zu erkennen, die mit dem CMS-Subtyp assoziiert sind. 

All dies führt zu einem tieferen Verständnis der DK-Klassifikation und -Prognose aus Sicht der komputationellen, experimentellen und klinischen Perspektive. 
Direct link to Lay Summary Last update: 24.07.2020

Responsible applicant and co-applicants


Project partner

Associated projects

Number Title Start Funding scheme
197786 Function and regulation of CDX2 in colon cancer tumor budding and dissemination 01.01.2021 Project funding
194809 Extending the Human (C)hemotype Project 01.09.2020 SNSF Professorships
127325 The evolution of proteins with tandem repeats: a large-scale study of rates, selective forces, complex patterns and associations with human disease 01.08.2010 Project funding
174836 C16.0072: Discovering evolutionary innovations by assessing variation and natural selection in protein tandem repeats 01.01.2017 COST (European Cooperation in Science and Technology)
127325 The evolution of proteins with tandem repeats: a large-scale study of rates, selective forces, complex patterns and associations with human disease 01.08.2010 Project funding


In 2018, colorectal cancer (CRC) was the third most frequently diagnosed cancer in Switzerland. Despite decades of research, 5-year survival for CRC patients is only 60%, and there exist few molecular biomarkers and treatment options. The Consensus Molecular Subtypes (CMS) published in 2015 provided the first comprehensive molecular classification of CRC with clinical implications (prognosis and treatment response prediction), but some aspects are missing. Here we aim to improve CRC patient stratification, prognosis and treatment prediction using an integrated approach to CRC trans-omic data.Microsatellite and STR variants are reported to be overrepresented in CRC tumors and are known to be associated with gene expression changes, however they are poorly studied due to technical limitations. Here, we will analyse genomics and other -omics data to systematically search for STR biomarkers, annotating and cataloguing tumor STR variants that confer a selective advantage to tumorigenesis of colorectal cancer. To this purpose, we will develop statistical methods to identify genetic markers displaying the adaptive signature of diversifying selection, commonly affecting tumor cells. This will be done using a newly developed evolutionary modeling approach combined with expression analyses and a genome-wide association study (GWAS) for the first time applied to STRs. The inferred CRC “risk” variants will be validated as novel RNA and protein targets using clinical samples, for which we will sequence full genomes and perform targeted mass-spectrometry proteomics experiments for risk loci.Next, taking a histomics approach, we will compare the performance of different machine learning algorithms and graph representations for prediction of CMS groups from histopathological slides to understand the distribution and heterogeneity of CMS in different clinical scenarios. We will evaluate the intratumoral and inter-tumoral heterogeneity to determine the relevance on outcome. In other words, we will use the image of each cancer as a surrogate for the CMS groups. These patterns may reflect tumors either with DNA mismatch repair defects (CMS1), WNT deregulation (CMS2), metabolic tumors (CMS3), or those with epithelial-mesenchymal transition-like (EMT) phenotype/cancer stem cell (CMS4). We will exploit not only publicly available datasets but will tap into our own rich repository of more than 300’000 CRC images and different cohorts with clinicopathological information. Together with our project partners, we will use different machine learning techniques to test the performance of different classifiers for CMS group prediction and their clinical relevance. We will also employ pharmacogenomics approaches, and develop an experimental platform to perform ex vivo image-based drug screening on patient biopsies from liver metastases of CRC patients enrolled into a recently approved clinical trial. We will evaluate and optimize the power of personal ex vivo drug responses to predict clinical response of individual patients. This will provide insights about molecular mechanisms within CRC cells and as a function of CMS class, identifying signaling pathways that play a role in responsiveness and resistance to therapeutic drugs, which potentially would help to identify new drug targets in the metastatic setting. Integration with patient-matched molecular measurements and biopsy-matched diagnostic H&E slides will provide insights into the molecular mechanisms underlying clinical response variability and enable further integration of the drug responses with the various technologies developed as part of this proposal.Finally, we will integrate all the available data types (genomics, proteomics and images) in a AI-driven multi-modal classifier and that will be trained to predict CMS classification, as well as other relevant clinical variables, such as staging, disease-free survival, etc. We will apply interpretability techniques to the deep learning classifier to extract insights about the molecular and morphological patterns associated with each CMS class. Evaluation of predictions on patient-matched molecular measurements and biopsy-matched diagnostic H&E slides will allow insights into the molecular mechanisms underlying clinical response variability and enable further integration with the various technologies developed as part of this proposal. Finally, we will adapt the multi-modal classifier to predict drug sensitivity on patient-specific molecular profiles. The four omics subprojects, while each having independent research agenda, are tightly interconnected. Success of the overall project will be ensured by interdisciplinary interactions, explicitly defined as work packages.