MetaGxData: Clinically Annotated Breast, Ovarian and Pancreatic Cancer Datasets and their Use in Generating a Multi-Cancer Gene Signature

Deena M.a. Gendoo, Michael Zon, Vandana Sandhu, Venkata Manem, Natchar Ratanasirigulchai, Gregory M. Chen, Levi Waldron, Benjamin Haibe‐KainsbioRxiv 2016 · DOI

TL;DR

This paper introduces MetaGxData, a comprehensive, standardized compendium of transcriptomic and clinical data for breast, ovarian, and pancreatic cancers. By harmonizing this data, they enable robust multi-cancer analyses, leading to the development of the first gene signature that is prognostic across all three cancer types.

What Problem Does This Paper Address?

The vast amount of publicly available transcriptomic and clinical data for various cancer types is a goldmine for research, but it's often locked away in unharmonized formats and varying standards. This makes it incredibly difficult to perform large-scale, cross-cancer analyses that could reveal common biological mechanisms or lead to more generalizable prognostic markers. Specifically, existing resources allow for single-dataset analyses but lack the standardization needed for robust meta-analyses, hindering the development of reliable, multi-cancer gene signatures.

Methods

The core of this work is the development of the MetaGxData R package. This involved:

  • Data Acquisition and Curation: Gathering data from numerous public microarray datasets for breast, ovarian, and pancreatic cancers.
  • Standardization: Developing a unified nomenclature and standardized clinical, pathological, and survival metadata across all datasets and cancer types. This included creating semantic variables for categorical data (e.g., ER status) and ensuring consistency in how variables like stage and grade were represented.
  • Probe-gene Mapping: A crucial step to ensure that gene expression values were consistently mapped to specific genes, even across different microarray platforms. They selected the probe with the highest variance for each gene.
  • Duplicate Sample Identification: Implementing a method to identify and handle samples with highly correlated profiles (Spearman correlation > 0.98) to avoid redundancy.
  • Prognostic Analysis: Utilizing Cox proportional hazards models and meta-analysis techniques (random-effects models) to estimate hazard ratios and identify prognostic genes and signatures.
  • Signature Development: Building a multi-cancer gene signature by identifying genes that were consistently prognostic across multiple datasets and cancer types.

Datasets

The study utilized a large collection of publicly available microarray datasets:

  • Breast Cancer: 39 datasets encompassing 10,004 samples.
  • Ovarian Cancer: 26 datasets encompassing 3,526 samples.
  • Pancreatic Cancer: 21 datasets encompassing 1,719 samples.

These datasets were integrated into the MetaGxBreast, MetaGxOvarian, and MetaGxPancreas R packages, which are publicly available through Bioconductor's ExperimentHub. This compendium represents the largest standardized collection of curated transcriptomic data for these cancer types to date.

Key Insights and Ideas

  • The Power of Harmonization: The paper strongly emphasizes that standardizing and integrating diverse datasets is not just an organizational task but a critical enabler of powerful cross-cancer research.
  • Shared Biology Across Cancers: The identification of common prognostic genes between breast and ovarian cancers highlights shared underlying biological processes that influence patient outcomes.
  • A Generalizable Multi-Cancer Signature: The development of a single gene signature that is prognostic across breast, ovarian, and pancreatic cancers is a significant achievement, suggesting that some molecular drivers of prognosis are conserved across these distinct malignancies.
  • Meta-Analysis as a Validation Engine: The study demonstrates how meta-analysis, powered by a standardized compendium, can provide more robust and generalizable results than analyses based on single datasets.

Key Results

  • MetaGxData Compendium: The creation of the largest standardized compendium of curated transcriptomic and clinical data for breast, ovarian, and pancreatic cancers, comprising 86 datasets and 15,249 samples.
  • Common Prognostic Genes: Identification of 30 genes that are significantly prognostic across both breast and ovarian cancers.
  • Multi-Cancer Gene Signature: Development of a novel gene signature that demonstrated prognostic value in a meta-analysis across breast, ovarian, and pancreatic cancers. This signature outperformed existing signatures in specific contexts.
  • Performance Benchmarking: The MetaGx signature was found to be more prognostic than clinically relevant breast cancer signatures (Mammaprint, Oncotype DX) in certain breast cancer subtypes and outperformed pancreas-specific signatures.
  • Robustness: The MetaGx signature's performance was significantly better than random signatures, indicating its biological relevance.

Limitations and Future Work

  • Data Heterogeneity: While efforts were made to standardize, the inherent heterogeneity of data from different platforms and studies remains a challenge.
  • Focus on Microarray Data: The current compendium primarily includes microarray data. Future work could expand to include RNA-seq data.
  • Validation Cohorts: While the study used validation cohorts for signature development, further independent validation in prospective clinical settings would be beneficial.
  • Expanding the Compendium: The authors envision MetaGxData as a living resource, with ongoing efforts to incorporate new datasets and potentially expand to other cancer types.
  • Deeper Mechanistic Studies: The identified multi-cancer signature provides a foundation for further investigation into the specific biological mechanisms driving prognosis across these cancers.

Summary updated

Abstract

ABSTRACT A wealth of transcriptomic and clinical data on solid tumours are under-utilized due to unharmonized data storage and format. We have developed the MetaGxData package compendium, which includes manually-curated and standardized clinical, pathological, survival, and treatment metadata across breast, ovarian, and pancreatic cancer data. MetaGxData is the largest compendium of curated transcriptomic data for these cancer types to date, spanning 86 datasets and encompassing 15,249 samples. Open access to standardized metadata across cancer types promotes use of their transcriptomic and clinical data in a variety of cross-tumour analyses, including identification of common biomarkers, establishing common patterns of co-expression networks, and assessing the validity of prognostic signatures. Here, we demonstrate that MetaGxData is a flexible framework that facilitates meta-analyses by using it to identify common prognostic genes in ovarian and breast cancer. Furthermore, we use the data compendium to create the first gene signature that is prognostic in a meta-analysis across 3 cancers. These findings demonstrate the potential of MetaGxData to serve as an important resource in oncology research and provide a foundation for future development of cancer-specific compendia.
BibTeX
@misc{gendoo2016metagxdata,
  author = {Deena M.A. Gendoo and Michael Zon and Vandana Sandhu and Venkata Manem and Natchar Ratanasirigulchai and Gregory M. Chen and Levi Waldron and Benjamin Haibe‐Kains},
  title = {{MetaGxData: Clinically Annotated Breast, Ovarian and Pancreatic Cancer Datasets and their Use in Generating a Multi-Cancer Gene Signature}},
  year = {2016},
  doi = {10.1101/052910},
  howpublished = {bioRxiv},
}
Figure