Computational approach for deriving cancer progression roadmaps from static sample data

Abstract

As with any biological process, cancer development is inherently dynamic. While major efforts continue to catalog the genomic events associated with human cancer, it remains difficult to interpret and extrapolate the accumulating data to provide insights into the dynamic aspects of the disease. Here, we present a computational strategy that enables the construction of a cancer progression model using static tumor sample data. The developed approach overcame many technical limitations of existing methods. Application of the approach to breast cancer data revealed a linear, branching model with two distinct trajectories for malignant progression. The validity of the constructed model was demonstrated in 27 independent breast cancer data sets, and through visualization of the data in the context of disease progression we were able to identify a number of potentially key molecular events in the advance of breast cancer to malignancy.

Molecular Profiles of Matched Primary and Metastatic Tumor Samples Support a Linear Evolutionary Model of Breast Cancer

“…Our findings support that molecular subtypes are not hardwired, and genotypes and phenotypes can shift over time [11], and show that cancer development follows common progression paths, as posited in [29]. We performed a series of interrogations that provided substantial support for the constructed model, and showed the utility of such a model for testing and generating hypotheses and providing novel insights into previous observations from the cancer-evolution perspective [28]. First, we demonstrated that the same progression pattern was repeatedly observed in additional 26 breast cancer datasets, which together with the METABRIC dataset comprise almost all of the breast tumor samples assayed over the past 15 years (> 9, 000 samples).…”

“…Given the human tumor sampling limitation, we previously devised a computational strategy [28] to derive pseudo time-series data from static samples (excised tissue specimens). The design is based on the idea that each sample can provide a snapshot of the disease process, and if the number of samples is sufficiently large we can recover a visualization of disease progression.…”

“…The pattern of progression was confirmed by analysis in a series of smaller independent breast cancer datasets, and the conceptual validity of the model was validated by the mapping of additional clinical and molecular data available from the METABRIC and TCGA data sets. Although these factors were not included in the mathematical modeling, DNA mutation rate and genomic instability were clearly associated with progression towards malignancy, and diseasespecific survival data aligned significantly with the paths to the most malignant model termini [28].…”

Deep Learning Approach to Identifying Breast Cancer Subtypes Using High-Dimensional Genomic Data

“…Human cancer is a heterogeneous disease initiated by random somatic mutations and driven by multiple genomic alterations (Hanahan and Weinberg, 2011;Sun et al, 2017). In order to move towards personalized patient treatment regimes, cancers of specific tissues have been divided into molecular subtypes based on the gene expression profiles of primary tumors (Sørlie et al, 2001(Sørlie et al, , 2003Curtis et al, 2012;Parker et al, 2009).…”

“…The desire for levels of accuracy that can ultimately lead to clinical utility continues to drive the field to refine breast cancer subtypes (Shen et al, 2013;Parker et al, 2009;Sun et al, 2017;Haibe-Kains et al, 2012;Sun et al, 2014) and to identify molecular subtypes in other cancers (Abeshouse et al, 2015;Cancer Genome Atlas Network, 2014). The recent establishment of international cancer genome consortia (Cancer Genome Atlas Network, 2012;Abeshouse et al, 2015;Cancer Genome Atlas Network, 2014;Curtis et al, 2012) has generally overcome the sample size issue.…”