Uniform genomic data analysis in the NCI Genomic Data Commons

Zhenyu Zhang, Kyle M. Hernandez, Jeremiah Savage, Shenglai Li, Daniel P. Miller, Stuti Agrawal, Francisco Ortuño, Louis M. Staudt, Allison P. Heath, Robert L. GrossmanNature Communications 2021 · DOI

TL;DR

The NCI's Genomic Data Commons (GDC) has developed and implemented a suite of standardized pipelines to process and harmonize diverse cancer genomics data from TCGA and TARGET programs. This effort aims to reduce batch effects and enable more robust, cross-project analyses for precision medicine research.

What Problem Does This Paper Address?

Cancer genomics research generates vast amounts of data from various sources and using different experimental and analytical methods. This heterogeneity leads to "batch effects," making it challenging to compare and integrate datasets from different studies or projects. The National Cancer Institute's Genomic Data Commons (GDC) was established to address this by creating a centralized repository of uniformly processed genomic and clinical data, facilitating data sharing and collaborative analysis for precision medicine. This paper details the pipelines and workflows developed by the GDC to achieve this harmonization.

Methods

The GDC employs a two-stage harmonization process:

  • Alignment: Raw sequencing reads are aligned to a single, unified human reference genome (GRCh38) using standardized pipelines for each data type. This ensures consistency in how reads are mapped.
  • Higher-Level Data Generation: Different pipelines then process these aligned reads to derive summary results, such as somatic mutations, gene expression levels, and copy number variations.

Key methodological aspects include:

  • Reference Genome: GRCh38 is used as the primary human reference genome, with the inclusion of human decoy sequences and viral decoy sequences to improve alignment accuracy and enable virus detection.
  • Somatic Variant Calling: Four distinct callers (MuTect2, VarScan2, MuSE, SomaticSniper) are used to identify somatic mutations, with results aggregated to leverage the strengths of each.
  • Data Harmonization Pipelines: Workflows are built using the Common Workflow Language (CWL) and containerized with Docker to ensure reproducibility and portability.
  • Data Integration: Techniques like t-distributed stochastic neighbor embedding (t-SNE) are used to visualize integrated genomic and epigenomic data across different cancer types.

Datasets

The primary datasets utilized and processed by the GDC include:

  • The Cancer Genome Atlas (TCGA): A large collection of cancer genomic and clinical data.
  • Therapeutically Applicable Research to Generate Effective Therapies (TARGET): Another significant cancer genomics dataset.

The paper also references the use of:

  • Human GENCODE release version 22: For gene models in variant annotation and RNA-Seq alignment.
  • miRBase version 21: For miRNA-Seq annotation.
  • Viral sequences: Including Human Cytomegalovirus (CMV), Epstein-Barr virus (EBV), Hepatitis B virus (HBV), Hepatitis C virus (HCV), Human immunodeficiency virus (HIV), Human herpesvirus 8 (KSHV), Human T-lymphotropic virus 1 (HTLV-1), Merkel cell polyomavirus (MCV), Simian vacuolating virus 40 (SV40), and Human papillomavirus (HPV).

Key Insights and Ideas

  • The Power of Harmonization: Standardizing analysis pipelines across diverse datasets is crucial for enabling robust, large-scale cancer genomics research and precision medicine initiatives.
  • Leveraging Multiple Variant Callers: No single somatic variant caller is perfect. By using multiple callers and combining their outputs, the GDC can achieve a better balance of sensitivity and specificity.
  • Importance of Decoy Sequences: Including human and viral decoy sequences in the reference genome improves alignment accuracy and allows for the detection of viral contributions to cancer.
  • Reproducibility is Key: The use of CWL and Docker ensures that the GDC's analysis pipelines are reproducible and can be shared with the research community.
  • Integrated Analysis Reveals Patterns: Combining multiple data types (mRNA, miRNA, methylation, copy number) and applying dimensionality reduction techniques like t-SNE can reveal distinct patterns and relationships between cancer types that might not be apparent from single-modality analyses.

Key Results

  • By June 2016, the GDC had analyzed over 50,000 raw sequencing data inputs and generated more than 100,000 higher-level derived data files.
  • A significant portion (56.0%) of "clean" somatic variants were identified by all four GDC somatic callers, with 71.6% of validated TCGA variants recalled by all four pipelines.
  • Spearman correlation between GDC and TCGA mRNA expression data for shared samples and genes was high (average of 0.944 for sample-to-sample and 0.929 for gene-to-gene).
  • t-SNE clustering of TCGA data using integrated molecular features successfully separated most cancer types, with some interesting groupings (e.g., colon and rectum cancers clustering together).

Limitations and Future Work

  • Somatic Variant Caller Limitations: The paper acknowledges that somatic variant detection is an evolving field, and no single caller is universally superior. While combining callers improves recall, it might decrease sensitivity. Future work includes developing MAF files that merge results from multiple tools.
  • Germline Variant Masking: The process of creating open-access MAF files by removing potential germline variants is stringent and may inadvertently remove some real somatic variants, especially in regions of low sequencing coverage. Users are encouraged to explore controlled-access MAFs for a more comprehensive set of variants.
  • RNA-Seq Batch Correction: The GDC does not currently perform RNA expression batch corrections, which could be a limitation for cross-project comparisons where batch effects are significant.
  • DEXSeq Exon-Level Data: DEXSeq exon-level quantification results were not yet imported into the GDC Data Portal at the time of publication but are planned for a future release.
  • Future Work: The GDC plans to continue reprocessing data as new workflows and algorithms are developed, ensuring the repository remains up-to-date and leverages the latest advancements in genomic analysis. They also aim to expand their harmonized datasets to include data from future projects.

Summary updated

Abstract

The goal of the National Cancer Institute's (NCI's) Genomic Data Commons (GDC) is to provide the cancer research community with a data repository of uniformly processed genomic and associated clinical data that enables data sharing and collaborative analysis in the support of precision medicine. The initial GDC dataset include genomic, epigenomic, proteomic, clinical and other data from the NCI TCGA and TARGET programs. Data production for the GDC started in June, 2015 using an OpenStack-based private cloud. By June of 2016, the GDC had analyzed more than 50,000 raw sequencing data inputs, as well as multiple other data types. Using the latest human genome reference build GRCh38, the GDC generated a variety of data types from aligned reads to somatic mutations, gene expression, miRNA expression, DNA methylation status, and copy number variation. In this paper, we describe the pipelines and workflows used to process and harmonize the data in the GDC. The generated data, as well as the original input files from TCGA and TARGET, are available for download and exploratory analysis at the GDC Data Portal and Legacy Archive ( https://gdc.cancer.gov/ ).
BibTeX
@inproceedings{zhang2021uniform,
  author = {Zhenyu Zhang and Kyle M. Hernandez and Jeremiah Savage and Shenglai Li and Daniel P. Miller and Stuti Agrawal and Francisco Ortuño and Louis M. Staudt and Allison P. Heath and Robert L. Grossman},
  title = {{Uniform genomic data analysis in the NCI Genomic Data Commons}},
  booktitle = {Nature Communications},
  year = {2021},
  doi = {10.1038/s41467-021-21254-9},
}
Figure