Cancer Subtype Characterization

Identification and characterization of cancer subtypes for refined therapeutic targeting is an important step towards improved patient outcomes. Recent research has shown that integrated analysis of different molecular data types generated by the TCGA project can be used to discover subtypes and suggest molecular differences relevant for therapeutic approaches.

Large-scale data analysis systems developed to perform comprehensive analyses of TCGA data, such as the Firehose pipeline developed by the Broad Institute, have strong algorithmic capabilities and generate analysis results required for subtype identification and characterization. Visualization tools that enable biologists to integrate and explore these large sets of results are an important prerequisite for their efficient interpretation.

To address this, we have developed a new interactive visualization method that supports biologists in the identification and characterization of cancer subtypes. Our method is designed to integrate many TCGA data types, including expression, copy number, methylation, mutation, and clinical data.

The core concept of our approach is to visualize sample groupings (partitions of the set of all samples for a given tumor type) and the relationships between these groupings in a given cancer type. Groupings of samples are obtained for instance by clustering mRNA, microRNA or DNA methylation profiles. Groups can also be derived from copy number levels of a particular gene or gene mutation status, e.g., one group for “wild-type”, one for “mutated”.

In the visualization, groupings are represented as columns and the groups are represented as blocks in these columns. Relationships between groups are visualized as ribbons of varying width drawn between neighboring columns. Wide ribbons encode a high co-occurrence of samples in the connected groups, while the absence of ribbons between blocks indicates mutual exclusion. This encoding provides a straightforward and scalable overview of the consistency of group memberships of samples across different data types.

Since many groupings are derived from molecular profiling data, interactive heat maps can be visualized directly in the columns. The rows of the heat maps correspond to samples and columns correspond to the profiled molecules. Finally, our method also supports the visualization and comparison of such molecular profiling data for selected sample groups in the context of pathway maps to enable biologists to characterize functional differences between potential subtypes.

We have added an import module to Caleydo to load data directly from the output of a Firehose pipeline run. Ultimately, it will be possible to launch Caleydo preloaded with the data for a selected tumor type from a website using Java WebStart.

Comparison of the groupings in three datasets (miRNA, Methylation and mRNA)

Copy number status of three genes (CDKN2a, CDKN2B, EGFR), grouping of mRNA data and mapping to WNT signaling pathway.

Detailed view on WNT Signaling Pathway with an overlay of the expression of the fourth grouping.

You can also download the poster with more pictures.

If you are interested in receiving updates on the project, especially considering the availability for download, please leave your details and we will notify you