xena-gdc-etl 0.2.0

Scripts for importing GDC data into UCSC Xena.

Extract, transform and load GDC data onto UCSC Xena.

Table of Contents

• Dependencies

• Installation

• Basic usage with command line tools

• Advanced usage with XenaDataset and its subclasses

• GDC ETL settings

• Documentation

Dependencies

Specific versions mentioned below have been tested. Eariler versions may still work but not guaranteed.

1. Python 2.7, 3.5+

   This pipeline has been tested with python 2.7, 3.5, 3.6 and 3.7. It may also work with other python 3 versions since it was originally designed to be single-source Python 2/3 compatible.

2. Requests v1.2.3

3. Numpy v1.15.0

4. Pandas v0.23.2

5. Jinja2 v2.10.1: used for generating metadata JSON.

6. lxml v4.2.0: used for parsing TCGA phenotype data

7. xlrd v1.1.0: used for reading TARGET phenotype data

Installation

• First clone the repository from GitHub by running git clone https://github.com/ucscXena/xena-GDC-ETL.git. Now, cd into xena-GDC-ETL directory and install the package using pip: pip install .

  If you are developing the package, you can use pip’s edit mode for installation: pip install -e ..

• You can also directly use pip to install the package. To get the latest code from GitHub master branch, run: pip install git+https://github.com/ucscXena/xena-GDC-ETL. To get the latest stable version, run: pip install xena-GDC-ETL.

• Dependencies can be installed either before or after cloning this repository. You can install them by running pip install -r requirements.txt.

• In general,

  • gdc.py contains functionalities related to GDC API, which requires no other modules in this package;

  • xena_dataset.py contains core functionalities for importing data from GDC to Xena and needs the gdc.py module in this package;

  • gdc2xena.py defines a command line tool which requires both gdc.py module and xena_dataset.py module in this package;

  • gdc_check_new.py defines a command line tool which requiresthe gdc.py module in this package.

Basic usage with command line tools

• Import selected project(s) and selected type(s) of data from GDC to Xena

  xge etl [-h] [-r ROOT]
          [-p PROJECTS [PROJECTS ...] | -P NOT_PROJECTS [NOT_PROJECTS ...]]
          [-t DATATYPE [DATATYPE ...] | -T NOT_DATATYPE [NOT_DATATYPE ...]]
          [-D DELETE]

  This tool will perform a full import of dataset(s) into the root directory (specified by the -r option) with a default directory tree. In general, a full import has 3 steps: downloading raw data, making Xena matrix from raw data and generating matrix associated metadata. Data from each step will be saved to corresponding directories, whose structure is like this:

  root
  └── projects
      ├── "Raw_Data"
      │   └── xena_dtype
      │       ├── data1
      │       ├── data2
      │       ├── ...
      │       └── dataN
      └── "Xena_Matrices"
          ├── projects.xena_dtype(1).tsv
          ├── projects.xena_dtype(1).tsv.json
          ├── projects.xena_dtype(2).tsv
          ├── projects.xena_dtype(2).tsv.json
          ├── ...
          ├── projects.xena_dtype(N).tsv
          └── projects.xena_dtype(N).tsv.json

  A dataset is defined by its project and data type. Projects of interest are provided through -p or -P option, and data types of interest are provided through the -t or -T option. Multiple inputs separated by whitespace are allowed and will be treated separately with all possible combinations. Valid projects should be valid project_id on GDC. Valid data types includes (without quotation marks): ‘htseq_counts’, ‘htseq_fpkm’, ‘htseq_fpkm-uq’, ‘mirna’, ‘masked_cnv’, ‘muse_snv’, ‘mutect2_snv’, ‘somaticsniper_snv’, ‘varscan2_snv’, ‘phenotype’, and ‘survival’. Upper case options (-P or -T) are mutually exclusive with corresponding lower case options, and they are used to define datasets of interest by excluding selections from either all projects on GDC or all supported data types. For example, the following command line imports 3 types of RNA-seq data for all but FM-AD projects from GDC to /home/user/xena_root:

  mkdir -p /home/user/xena_root
  xge etl -P FM-AD -t htseq_counts htseq_fpkm htseq_fpkm-uq

  Notes:

  1. Root directory must be existing

  2. Please check the next section for advanced usage with XenaDataset and its subclasses, if you want to customize the importing process with selected (rather than all possible) combinations of your input projects and data types or selected (rather than all 3) importing step(s).

• Generate metadata of a xena matrix

  xge metadata --project TCGA-BRCA --datatype htseq_counts --matrix path/to/matrix.tsv --release 10

  This tool generates metadata for a xena matrix. For the shown example, metadata is generated for the matrix matrix.tsv for release 10, project TCGA-BRCA and datatype htsep_counts. Note that, metadata JSON file is saved at the same directory as the matrix.tsv file.

• Check against a list of updated files for affected dataset(s)

  xge gdc_check_new [-h] URL

  This tool takes in a file (either a URL or a local file readable by pandas.read_csv) of table and read one of its columns named as “New File UUID”. It then checks all file UUIDs in this table on GDC and summarize all their associated project(s), data type(s) and analysis workflow type(s). Such tables are usually provided in GDC’s data release note. With the summarized info, you can design specific imports to just update datasets which are updated on GDC. For example, the following command:

  xge gdc_check_new https://docs.gdc.cancer.gov/Data/Release_Notes/DR9.0_files_swap.txt.gz

  should give you:

  analysis.workflow_type    cases.project.project_id    data_type
  HTSeq - FPKM              TARGET-NBL                  Gene Expression Quantification
  HTSeq - FPKM-UQ           TARGET-NBL                  Gene Expression Quantification
  HTSeq - Counts            TARGET-NBL                  Gene Expression Quantification

• Shows the current version of xena_gdc_etl

  xge --version

• Check equality of two xena matrices

  xge xena-eql path/to/matrix1.tsv path/to/matrix2.tsv

  This tool takes path to two xena matrices and output if they are equal or not.

• Merge xena matrices

  xge merge-xena -f path/to/matrix1.tsv path/to/matrix2.tsv -t htseq_counts -o path/to/output -n new_name.tsv -c TCGA-BRCA

  This tool merges xena matrices and outputs the merged matrix. For the given example the tool will merge matrix1.tsv and matrix2.tsv matrices and store the merged matrix in path/to/output directory with the name new_name.tsv. Note that, had the argument -n not been specified, the merged matrix would have been saved as TCGA-BRCA.htseq_counts.tsv.

Advanced usage with XenaDataset and its subclasses

• The XenaDataset class

  Though this is not an abstract class, it is designed as a generalized class representing one Xena dataset and its importing process. For doing an import of GDC data, use its subclasses, which have preloaded with some default settings, might be simpler.

  A Xena dataset is defined by its study project (cohort) and the type of data in this dataset. A typical importing process has the following 3 steps:

  1. Download raw data from the source.

  The download_map property defines a dict of raw data to be downloaded, with the key being the URL and the value being the path, including the filename, for saving corresponding downloaded file. The download method will read the download_map and perform the downloading, creating non-existing directories as needed. After downloading all files, a list of paths for downloaded files will be recorded in the raw_data_list property. The download method needs only a valid download_map. It will return the object itself, therefore can be chained with transform.

  2. Transform raw data into valid Xena matrix.

  One assumption for data transformation is that there might be multiple raw data (in the raw_data_list) supporting the single Xena matrix in a dataset. Therefore, the transform method will first merge data and then process merged matrix as needed. It will open the file one by one accordingly (by extension), and read the file object and transform its data with a function defined by read_raw. The list of transformed single data will be merged and processed by a function defined by raws2matrix, which gives the finalized Xena matrix. The transform method requires a valid list of raw data, besides read_raw and raws2matrix. A valid list of raw data can be either explicitly defined by raw_data_list or can be derived from raw_data_dir with all files under raw_data_dir being treated as raw data. It will return the object itself, therefore can be chained with metadata.

  3. Generate metadata for the new Xena matrix.

  Metadata for Xena matrix is a JSON file rendered by the metadata method with metadata_vars (dict) through Jinja2 from metadata_template. This JSON file will be saved under the same directory as the matrix, with a filename being the matrix name plus the ‘.json’ postfix. The metadata method requires an existing file of Xena matrix.

  root_dir is both an optional instantiation arguments and a property. By default, it points to the current working directory. It is worth mentioning that the default directory structure mentioned above is implemented in the class. However, you are free to changed the setting with the following properties:

  • Pass an argument for root_dir during instantiation or set the root_dir property explicitly after instantiation.

  • Downloaded raw data will be saved under raw_data_dir.

  • Newly transformed Xena matrix will be saved as matrix under matrix_dir. The directory path in matrix has the priority over matrix_dir. By default, Xena matrix will be saved under the “matrix_dir” as “<projects>.<xena_dtype>.tsv”.

  • Metadata will always have the specific pattern of name and be together with matrix (i.e. no way to change this behavior).

• Build GDC importing pipelines with GDCOmicset, GDCPhenoset or GDCSurvivalset classes

  GDCOmicset, GDCPhenoset and GDCSurvivalset are subclasses of XenaDataset and are preloaded with settings for importing GDC genomic data, TCGA phenotype data on GDC, TARGET phenotype data on GDC and GDC’s survival data respecitively. These settings can be customized by setting corresponding properties described below. For more details, please check the next section and the documentation.

  The script for gdc2xena.py command line is a good example for basic usage of these classes. Similar to XenaDataset, a GDC dataset is defined by projects, which is one or a list of valid GDC “project_id”. For GDCOmicset, a dataset should also be defined with one of the supported xena_dtype (find out with the class method GDCOmicset.get_supported_dtype()). The xena_dtype is critical for a GDCOmicset object selecting correct default settings. For GDCPhenoset and GDCSurvivalset, data type are self-explanatory and cannot be changed. Therefore, you can instantiate these classes like this:

  from xena_dataset import GDCOmicset, GDCPhenoset, GDCSurvivalset, GDCAPIPhenoset
  
  gdc_omic_cohort = GDCOmicset('TCGA-BRCA', 'htsep_counts')
  
  # Won't check if the ID is of TCGA program or not.
  tcga_pheno_cohort = GDCPhenoset('TCGA-BRCA')
  
  # Won't check if the ID is of TARGET program or not.
  target_pheno_cohort = GDCPhenoset('TARGET-NBL')
  
  gdc_survival_cohort = GDCSurvivalset('TCGA-BRCA')
  
  gdc_api_pheno_cohort = GDCAPIPhenoset('CPTAC-3')

  With such a dataset object, it is fine to call download, transform and/or metadata method(s). These methods will use preloaded settings and save files under root_dir accordingly. You are free to call/chain some but not all 3 methods; just keep in mind the pre-requisites for each method and set related properties properly. Aside from directory related settings described above, you can change some default importing settings through the following properties.

  • Customize GDCOmicset

  
  

  Attributes	Usage	Type and Format1	Default settings
  gdc_filter	Used for deriving default download_map as the GDC search filters.	dict: the key is 1 GDC available file field and the value is either a string or a list, meaning the value of the file field matches a string or number in (a list)	Check GDC download settings for details.
  gdc_prefix	Used for deriving default download_map as the GDC search fields.	str: 1 GDC available file field whose value will be the prefix of the filename of corresponding downloaded file.	Check GDC download settings for details.
  download_map	Used by the download method for downloading GDC raw data supporting this dataset.	dict: the key is download URL and the value is the desired path for saving the downloaded file.	Download URLs are in the pattern of “https://api.gdc.cancer.gov/data/<FILE UUID>”, and paths are in the pattern of “<raw_data_dir>/<value of gdc_prefix>.<GDC file UUID>.<file extension>”.
  read_raw	Used by the transform method when reading a single GDC raw data.	callable: takes only 1 file object as its argument and returns an arbitrary result which will be put in a list and passed on to raws2matrix.	Check GDC genomic transform settings for details
  raws2matrix	Used by the transform method and responsible for both merging multiple GDC raw data into one Xena matrix and processing new Xena matrix as needed.	callable: takes only 1 list of read_raw returns as its argument and returns an object (usually a pandas DataFrame) which has a to_csv method for saving as a file.	Check GDC genomic transform settings for details
  metadata_template	Used by the metadata method for rendering metadata by Jinja2.	jinja2.environment.Template or str: a jinja2.environment.Template used directly by Jinja2; if it’s a string, it is a path to the template file which will be silently read and converted to jinja2.environment.Template.	Resources
  metadata_vars	Used by the metadata method for rendering metadata by Jinja2.	dict: used directly by Jinja2 which should match variables in metadata_template.	{ 'project_id': <``projects``>, 'date': <the time of last modification of ``matrix``>, 'gdc_release': <``gdc_release``>, 'xena_cohort': <Xena specific cohort name for TCGA data or GDC project_id for TARGET data, with (for both) "GDC " prefix> } * The first element of the “url” field in metadata will be “gdc_release” URL, and the second will be specific URL for raw data file if there is only 1 raw data file for this dataset; or it will be just “https://api.gdc.cancer.gov/data/”.
  gdc_release	Used by the metadata method for rendering metadata, showing the GDC data release of this dataset.	str: an URL pointing to corresponding GDC Data Release Note.	Current data release version when the gdc_release is being used/called, queried through “https://api.gdc.cancer.gov/status”.

  1. GDC API Available File Fields: https://docs.gdc.cancer.gov/API/Users_Guide/Appendix_A_Available_Fields/#file-fields

  • Customize GDCPhenoset for TCGA projects

    TCGA phenotype data for Xena includes both clinical data and biospecimen data, as detailed below. Downloading and transformation of clinical data and biospecimen data are in fact delegated by two independent GDCOmicset object respecitively. Corresponding subdatasets can be accessed through clin_dataset and bio_dataset attributes and hence can be customized as mentioned above. Because of such complexity of TCGA phenotype data, the download and transform methods are coded specifically and overrode corresponding methods of the base class, XenaDataset. Customization for downloading and matrix transformation is very limited and should be done in the following steps:

    1. Instantiate a GDCPhenoset;

    2. Instantiate and customize one or two GDCOmicset objects for clinical data and/or biospecimen data as needed;

    3. Assign customized GDCOmicset objects to corresponding attributes, clin_dataset and bio_dataset;

    4. Call desired method(s) (download and/or transform).

    • Customize download step

      This step can be customized only through customized clin_dataset and bio_dataset, since the whole downloading process is delegated by these two GDCOmicset objects.

    • Customize transform step

      The first part of transform is delegated by transform methods of clin_dataset and bio_dataset. Therefore, the only way to customized this process is to customize clin_dataset and bio_dataset. How the two matrices are then merged into one Xena phenotype matrix is hard coded and cannot be customized. It is worth noting that if you want to call transfrom but skip the downloading step, you will need to define clin_dataset and bio_dataset before calling transform.

    • Customize metadata step

      Different from download and transform, there is no special settings for the metadata method of GDCPhenoset. Therefore, similar to that of GDCOmicset, this step can be customized through metadata_template, metadata_vars and gdc_release properties. And to call just the metadata method, an existing matrix is enough.

  • Customize GDCPhenoset for TARGET projects

    TARGET phenotype data for Xena contains only the clinical data (no biospecimen data), as detailed below. The importing process is quite similar to that of a GDCOmicset. You can customize TARGETPhenoset with download_map, read_raw, raws2matrix, metadata_template, metadata_vars and gdc_release in the same way as that of GDCOmicset.

  • Customize GDCSurvivalset

    GDC data supporting Xena survival matrix does not come any GDC files. It comes from the “analysis/survival” endpoint of GDC API. Therefore, the download and transform methods are re-designed, overriding those of the base class, XenaDataset. Aside from redefining download and transform methods, there is no simple way to customize download and transform steps. You can still call transform without download by just defining a valid list of raw data with raw_data_list or raw_data_dir. However, only this first file in the list will be read and used.

    Different from download and transform, there is no special settings for the metadata method of GDCSurvivalset. Therefore, similar to that of GDCOmicset, this step can be customized through metadata_template, metadata_vars and gdc_release properties. To call just the metadata method, an existing matrix is enough.

  • Customize GDCAPIPhenoset

    The data for this class comes from GDC API only. Therefore, the download and transform methods are re-designed, overriding those of the base class, XenaDataset. Aside from redefining download and transform methods, there is no simple way to customize download and transform steps.

    Different from download and transform, there is no special settings for the metadata method of GDCAPIPhenoset. Therefore, similar to that of GDCOmicset, this step can be customized through metadata_template, metadata_vars and gdc_release properties. To call just the metadata method, an existing matrix is enough.

GDC ETL settings

• Settings for downloading/getting raw data (files) from GDC

  xena_dtype	GDC API endpoint	GDC data filter	File count/Level	GDC file field for filename prefix
  data_type	analysis.workflow_type
  htseq_counts	data	Gene Expression Quantification	HTSeq - Counts	1/Sample vial	cases.samples.submitter_id
  htseq_fpkm	data	Gene Expression Quantification	HTSeq - FPKM	1/Sample vial	cases.samples.submitter_id
  htseq_fpkm-uq	data	Gene Expression Quantification	HTSeq - FPKM-UQ	1/Sample vial	cases.samples.submitter_id
  mirna	data	miRNA Expression Quantification	BCGSC miRNA Profiling	1/Sample vial	cases.samples.submitter_id
  mirna_isoform	data	Isoform Expression Quantification	BCGSC miRNA Profiling	1/Sample vial	cases.samples.submitter_id
  cnv	data	Copy Number Segment	DNAcopy	1/Sample vial	cases.samples.submitter_id
  masked_cnv	data	Masked Copy Number Segment	DNAcopy	1/Sample vial	cases.samples.submitter_id
  gistic	data	Gene Level Copy Number Scores	GISTIC - Copy Number Score	1/Project	submitter_id
  star_counts	data	STARCounts	STAR - Counts	1/Sample vial	cases.samples.submitter_id
  muse_snv	data	Masked Somatic Mutation	MuSE Variant Aggregation and Masking	1/Project	submitter_id
  mutect2_snv	data	Masked Somatic Mutation	MuTect2 Variant Aggregation and Masking	1/Project	submitter_id
  somaticsniper_snv	data	Masked Somatic Mutation	SomaticSniper Variant Aggregation and Masking	1/Project	submitter_id
  varscan2_snv	data	Masked Somatic Mutation	VarScan2 Variant Aggregation and Masking	1/Project	submitter_id
  clinical	data	Clinical Supplement	N/A	0 or 1/Case	cases.submitter_id
  biospecimen	data	Biospecimen Supplement	N/A	1/Case	cases.submitter_id
  survival	analysis/survival	N/A (filtered by just the “project.project_id”)		1 Record/Case (Non-file)	N/A (filename will be “<projects>.GDC_survival.tsv”)

• Settings for transform “Omic” data into Xena matrix

  xena_dtype	Raw data has header?	Select columns (in order)	Row index	Skip rows start with?	Merge into matrix as	Process matrix
  htseq_counts	No	1, 2 [Ensembl_ID, Counts]	Ensembl_ID	_	1 new column based on index	Average if there are multiple data from the same sample vial; log2(counts + 1)
  htseq_fpkm	No	1, 2 [Ensembl_ID, Counts]	Ensembl_ID	_	1 new column based on index	Average if there are multiple data from the same sample vial; log2(counts + 1)
  htseq_fpkm-uq	No	1, 2 [Ensembl_ID, Counts]	Ensembl_ID	_	1 new column based on index	Average if there are multiple data from the same sample vial; log2(counts + 1)
  mirna	Yes	1, 3 [miRNA_ID, RPM]	miRNA_ID	N/A	1 new column based on index	Average if there are multiple data from the same sample vial; log2(counts + 1)
  mirna_isoform	Yes	2, 4 [isoform_coords, RPM]	isoform_coords	N/A	1 new column based on index	Average if there are multiple data from the same sample vial; log2(counts + 1)
  cnv	Yes	2, 3, 4, 6 [Chromosome, Start, End, Segment_Mean]	sample	N/A	New rows based on column name	Rename columns as: { 'Chromosome': 'Chrom', 'Segment_Mean': 'value' }
  masked_cnv	Yes	1, 2, 3, 5 [Chromosome, Start, End, Segment_Mean]	sample	N/A	New rows based on column name	Rename columns as: { 'Chromosome': 'Chrom', 'Segment_Mean': 'value' }
  gistic	Yes	1 [Ensembl_ID]	Ensembl_ID	_	N/A	Drop “Gene ID” and “Cytoband” column; Map “samples.portions.analytes.aliquots.aliquot_id” into “samples.submitter_id” using GDC API and use it as index.
  muse_snv mutect2_snv somaticsniper_snv varscan2_snv	Yes	13, 37, 5, 6, 7, 40, 42, 52, 1, 11, 16, 111 [Tumor_Seq_Allele2, HGVSp_Short, Chromosome, Start_Position, End_Position, t_depth, t_alt_count, Consequence, Hugo_Symbol, Reference_Allele, Tumor_Sample_Barcode, FILTER]	N/A	#	N/A	Calculate variant allele frequency (dna_vaf) by “t_alt_count”/”t_depth”; Delete “t_alt_count” and “t_depth” columns; Trim “Tumor_Sample_Barcode” to sample vial level; Rename columns as: { 'Hugo_Symbol': 'gene', 'Chromosome': 'chrom', 'Start_Position': 'start', 'End_Position': 'end', 'Reference_Allele': 'ref', 'Tumor_Seq_Allele2': 'alt', 'Tumor_Sample_Barcode': 'sampleid', 'HGVSp_Short': 'Amino_Acid_Change', 'Consequence': 'effect', 'FILTER': 'filter' }
  star_counts	Yes	1, 2 [Ensembl_ID, Counts]	Ensembl_ID	_	1 new column based on index	Average if there are multiple data from the same sample vial; log2(counts + 1)

• Settings for transform phenotype data into Xena matrix

  GDC program	GDC raw data	Raw data format	Single data file transformation	Merge and matrix processing
  TCGA	Clinical Supplement and Biospecimen Supplement	BCR XML	For clincial data, info will be extracted and organized into a per patient based pandas DataFrame. It will have a column named “bcr_patient_barcode” which will be used to join with biospecimen matrix later on. The XML scheme are quite different for different projects. Therefore, to get as much info as possible while still keeping things clear, texts, if any, from all elements that have non-element children are extracted first. After such a “dirty” extraction, two clean ups will be done: For “race” info, it will be converted into a comma separated list of races, in case there are more than one entry in <clin_shared:race_list> in the clinical XML. When there is one or more follow ups, the most recent follow up will be find out. All info in the most recent follow up will be used to replace/add to previously extracted matrix. For biospecimen data, there is one coherent XML scheme for all TCGA projects. There are two parts to be considered for biospecimen data: per sample/sample specific data and patient data (which is common for all samples). Info from both parts will be extracted and finally organized into a per sample based matrix, having a column named “bcr_patient_barcode”, which will be used to join with clinical matrxi later on. In general, info extraction has the following 3 steps: Common patient data will be extracted first, including texts from direct children of <admin:admin> and <bio:patient>. A new field of “primary_diagnosis” will be added by mapping “disease_code” to TCGA study name. Samples from <bio:patient/bio:samples> will be processed and have comman patient data attached one by one. Non-empty texts from direct children of sample will be extracted, i.e. details from nodes like <bio:portions> will be dropped. Samples having type code 10 are dropped. A column of “bcr_patient_barcode” from <bio:patient/shared:bcr_patient_barcode> will be added to the final biospecimen matrix (same for the whole table).	Multiple clinical data are concatenated directly by row with all empty columns removed. Multiple biospecimen data are concatenated directly by row with all empty columns removed. Merged clinical matrix and merged biospecimen matrix are further merged on “bcr_patient_barcode”. For conflict/overlapping columns, non-empty value from the clinical data has the priority.
  TARGET	Clinical Supplement only	XLSX	The excel file is converted to a pandas DataFrame.	Multiple DataFrames will be concatenated directly by row, and arriage return and line feed are replaced by a single space. Clinical data is per case(patient) based, while Xena phenotype matrix is per sample based. All related samples for each case/patient will be identified and phenotype data will be mapped to corresponding samples.

• Settings for transform survival data into Xena matrix

  GDC survival data is returned as JSON from GDC API. During the downloading process, it can and will be converted directly to pandas DataFrame and saved as tab delimited table. During transformation, columns in “primary” Xena survival matrix can be mapped directly (without further processing/calculation) from the raw table like this:

  Primary Xena column	GDC source column
  OS.time	time
  OS	censored
  _PATIENT	submitter_id

  GDC survival data is per case(patient) based and so is “primary” Xena survival matrix, while Xena survival matrix is per sample based. All related samples for each case/patient will be identified and survival data will be mapped to corresponding samples.

• CPTAC-3 Cohort

  CPTAC-3 data consists of RNAseq data (as discussed in GDCOmicset) and clinical data from the API. The cases and expand for clinical data are defined in the constants.py file.

Documentation

Check documentation for GDC module and Xena Dataset module here.