Talk Title:

Cancer Data Science: Mutation, Selection and Computation

Talk Summary:

The Gerstung lab develops statistical algorithms to learn more about the causes and consequences of cancer from large data sets. I will provide an overview about recent advances in the areas of cancer evolution, mutational signature analysis, as well as digital pathology and spatial genomics. Our analyses of the large cohort of cancer genomes assembled by the PCAWG consortium revealed that cancer driver mutations precede diagnosis by years to decades. These time scales are also observed in precancer data sets and provide a time frame within which transformation to malignant cancer can be predicted from genomic data. I’ll further present algorithms for learning intragenomic variation of mutational signatures and data from a large mutagenesis screen detailing how the interwoven and often counteracting forces of DNA damage, tolerance and repair jointly sculpt mutational signatures. Lastly I’ll present some of the labs forays into digital pathology, which reveal that a broad range of genomic and transcriptomic alterations are associated with histopathological characteristics which can be automatically learned with deep learning and subsequently localised within large tumor sections. Such observations are further corroborated by new spatial genomics and transcriptomics technologies that reveal subclonal growth patterns and their associated molecular and histopathological characteristics.