An automated data integration platform for interpreting genomic data and reporting treatment options in molecular tumor boards

by Jingyu Yang
Doctoral thesis
Date of Examination:2024-08-12
Date of issue:2024-08-16
Advisor:Prof. Dr. Tim Beißbarth
Referee:Prof. Dr. Ulrich Sax
Referee:Prof. Dr. Matthias Dobbelstein
Referee:Prof. Dr. Anne-Christin Hauschild
Referee:Prof. Dr. Günter Schneider
Referee:PD Dr. Elisabeth Heßmann
Persistent Address: http://dx.doi.org/10.53846/goediss-10681

 

 

Files in this item

Name:Jingyu_Yang_Dissertation.pdf
Size:20.7Mb
Format:PDF

The following license files are associated with this item:

Abstract

English

This thesis presents significant advancements in precision oncology, focusing on enhancing MTB workflows through the development and integration of innovative bioinformatics tools and methodologies. These contributions address the critical challenges of processing and interpreting complex genomic data, which are pivotal for personalized cancer treatment decisions. The foundation of this work is the development of Onkopipe, an advanced NGS bioinformatics pipeline that efficiently processes SNV, CNV, and SV without the need for matched normal samples. This pipeline facilitates a high-throughput, accurate genomic profiling essential for identifying actionable cancer mutations. Its containerized architecture ensures consistent and replicable analysis across different computational environments, making it an invaluable tool for both clinical and research settings. Standard reference data HD753 and NA12868 were used for pipeline validation. Building upon this foundation, the MTB-Report tool is introduced to annotate, filter, and sort genetic variants using data from public databases, providing a dual interface for both real-time clinical settings and batch processing for research applications. This tool enhances the MTB's ability to make informed treatment decisions based on a solid evidence base. Further enhancing the MTB platform arsenal, Onko_DrugCombScreen, a Shiny-based application, is developed to identify optimal drug combinations by integrating patient genomic data with extensive drug knowledge bases. This tool addresses the complexity of drug combination discovery by providing evidence-backed, patient-specific treatment recommendations, thus optimizing therapeutic strategies in precision oncology. A case study based on TCGA BRCA data, along with the drug combinations approved and currently undergoing clinical trials in breast cancer subtypes, affirms our approach. Additionally, an innovative method was explored using GCNN combined with LRP to interpret the BRCAness phenotype—a key aspect of precision oncology associated with homologous recombination repair deficiencies and sensitivity to PARP inhibitors. This computational approach reveals critical genes and pathways of BRCAness, providing deeper insights into the genetic basis of cancer subtypes and their potential therapeutic targets. In summary, these tools and methodologies collectively construct an integrated platform solution that simplifies the analysis, integration, interpretation, and application of genomic and clinical data, enhancing the efficiency and effectiveness of Molecular Tumor Boards and strengthening the rationality of precision medication. This work not only bridges the gap between genomic data and clinical application but also lays the foundation for the future development of the field of precision oncology.
Keywords: Precision and Personalized medicine; Explainable Artificial Intelligence; Graph Convolutional Neural Network; Layer-wise Relevance Propagation; Molecular tumor board; Drug combination; Next Generation Sequencing
 
Statistik