csr2transmart 0.4.0

Script to load CSR data to TranSMART

This package contains a script that transforms Central Subject Registry data to a format that can be loaded into the TranSMART platform, an open source data sharing and analytics platform for translational biomedical research.

The output of the transformation is a collection of tab-separated files that can be loaded into a TranSMART database using the transmart-copy tool.

⚠️ Note: this is a very preliminary version, still under development. Issues can be reported at https://github.com/thehyve/python_csr2transmart/issues.

Installation and usage

To install csr2transmart, do:

pip install csr2transmart

or from sources:

git clone https://github.com/thehyve/python_csr2transmart.git
cd python_csr2transmart
pip install .

Data model

The Central Subject Registry (CSR) data model contains individual, diagnosis, biosource and biomaterial entities. The data model is defined as a data class in csr/csr.py

To learn how to add changes to the database model, see changing-data-model.rst document.

Usage

This repository contains a number of command line tools:

• sources2csr: Reads from source files and produces tab delimited CSR files.

• csr2transmart: Reads CSR files and transforms the data to the TranSMART data model, creating files that can be imported to TranSMART using transmart-copy.

• csr2cbioportal: Reads CSR files and transforms the data to patient and sample files to imported into cBioPortal.

sources2csr

sources2csr <input_dir> <output_dir> <config_dir>

The tool reads input files from <input_dir> and writes CSR files in tab-delimited format (one file per entity type) to <output_dir>. The output directory <output_dir> needs to be either empty or not yet existing.

The sources configuration will be read from <config_dir>/sources_config.json, a JSON file that contains the following attributes:

• entities: a map from entity type name to a description of the sources for that entity type. E.g.,

  {
    "Individual": {
      "attributes": [
        {
          "name": "individual_id",
          "sources": [
            {
              "file": "individual.tsv",
              "column": "individual_id"
            }
          ]
        },
        {
          "name": "birth_date",
          "sources": [
            {
              "file": "individual.tsv",
              "date_format": "%d-%m-%Y"
            }
          ]
        }
      ]
    }
  }

  The entity type names have to match the entity type names in the CSR data model and the attribute names should match the attribute names in the data model as well. The column field is optional, by default the column name is assumed to be the same as the attribute name. For date fields, a date_format can be specified. If not specified, it is assumed to be %Y-%m-%d or any other date formats supported by Pydantic. If multiple input files are specified for an attribute, data for that attribute is read in that order, i.e., only if the first file has no data for an attribute for a specific entity, data for that attribute for that entity is read from the next file, etc.

• codebooks: a map from input file name to codebook file name, e.g., {"individual.tsv": "codebook.txt"}. Naming convention: <fileName>_codebook.txt

• file_format: a map from input file name to file format configuration, which allows you to configure the delimiter character (default: \t). E.g., {"individual.tsv": {"delimiter": ","}}.

See test_data/input_data/config/sources_config.json for an example.

Content of the codebook files has to match the following format:

• First a header line with a number and column names the codes apply to. The first field has a number, the second field a space separated list of column names, e.g., 1\tSEX GENDER.

• The lines following the header start with an empty field. Then the lines follow the format of code\tvalue until the end of the line, e.g., \t1\tMale\t2\tFemale.

• The start of a new header, which is detected by the first field not being empty starts the process over again.

See test_data/input_data/codebooks/valid_codebook.txt for a codebook file example.

csr2transmart

csr2transmart <input_dir> <output_dir> <config_dir>

The tool reads CSR files from <input_dir> (one file per entity type), transforms the CSR data to the TranSMART data model. In addition, if there is --ngs-dir specified, the tool will read the NGS files inside to determine values of additional CSR biomaterial variables. The tool writes the output in transmart-copy format to <output_dir>. The output directory <output_dir> needs to be either empty or not yet existing.

The ontology configuration will be read from <config_dir>/ontology_config.json. See test_data/input_data/config/ontology_config.json for an example.

csr2cbioportal

csr2cbioportal <input_dir> [--ngs-dir <ngs_dir>] <output_dir>

The tool reads CSR files from <input_dir> (one file per entity type), and optionally NGS data (genomics data) from <ngs_dir>, transforms the CSR data to the clinical data format for cBioPortal and writes the following data types to <output_dir>:

• Clinical data

• Mutation data

• CNA Segment data

• CNA Continuous data

• CNA Discrete data

File structure, case lists and meta files will also be added to the output folder. See the cBioPortal file formats documentation for further details.

The output directory <output_dir> needs to be either empty or not yet existing.

Source data assumptions and validation

General file characteristics

• Delimiter The source data should be provided as delimited text files. The delimiter can be configured per data file. If not configured, a tab-delimited file is assumed.

• Comments Comment lines may be present, indicated by a # as the first character. These lines will be ignored.

• Header The first non-comment line is assumed to be the header. It should be exactly one line.

• Field number The number of fields (columns) is determined by the header. Every other line in the file should have this same number of fields (no blank lines).

• Whitespace Leading or trailing whitespace is not trimmed. If present, it will persist in the final observation.

• Encoding All files are assumed to be utf-8 encoded.

CSR entities

All characteristics and relationships of the CSR data model are defined in csr/csr.py. Any field present in the source data that you would like to load to tranSMART, must be linked to a CSR field via the sources_config. Additional fields not present in the sources_config will be ignored.

Regarding the source data, we can distinguish four types of validation:

1. Value validation Independent validation of a single field value. This comprises type validation (e.g. string, integer or date), nullability (whether a field may be empty), and unique constraints.

2. Record validation Validation across different fields from the same record within the same entity. This validation is relevant when the validity of a field value is dependent on the other fields of the same record (e.g. a biosource record with src_biosource_id = BS1, is invalid when biosource_id = BS1).

3. Entity validation Concerns the integrity check of all records within a single entity (e.g. do all src_biosource_id values also have corresponding biosource_id records within the biosource entity).

4. Across-entity validation Checks the validity of relationships between records of different entities.

The data validation of the current pipeline is implemented for type 1 and to a limited extent for type 2 and 4. Hence, the source data is assumed to be coherent regarding its relationships within the same entity and across different entities. While most erroneous relationships across entities, in respect of missing entity records, will be detected (e.g. a biomaterial linked to a non-existing biosource), logically impossible relationships are not (e.g. biomaterial BM2 is derived from BM1, but from a different biosource).

All entities must have a link to an individual, either through their individual-referencing field or through a reference to an entity of other type that has an individual-referencing field. Otherwise an error will be thrown.

Additionally, any individual needs to have at least one observation to be included. This means that merely a collection of related ID values, without observations linked to any of those IDs, will not become available in tranSMART.

NGS data

All NGS data should follow reference genome HG38. The sample ID is structured as <BiosourceID>_<BiomaterialID>, based on the IDs of biosources and biomaterials from clinical data, e.g. PMCBS000AAA_PMCBM000AAA.

The naming conventions for NGS input data are as follows:

• mutation data (Small nucleotide variants): <fileName>.maf.gz

• Segment data: <fileName>.seg

• Continuous CNA per gene: <fileName>_all_data_by_genes.txt

• Discrete CNA per gene: <fileName>_all_thresholded.by_genes.txt

Multiple files per type are supported. Note that mutation files must be always compressed as maf.gz. See the NGS test data for an example.

Each of the NGS file names can contain the optional string ‘_WGS’ or ‘_WSX’ to indicate the NGS analysis type, e.g. <fileName>_WXS_all_data_by_genes.txt. If present, each sample in the NGS file will be associated with the corresponding value for the derived variable analysis_strategy. String mapping rules are encoded in ngs_reader.py.

Additionally, the derived variable library_strategy is generated for each sample based on the NGS file extension. See specific NGS reader scripts in sources2csr for mapping rules between file type and and library strategy.

Both derived variables are associated to the entity Biomaterial in TranSMART (see ngs2csr.py); to have them appear in the TranSMART ontology tree after data loading, you need to include them in ontology_config.json, e.g.:

{
  "name": "Biomaterial information",
  "children": [
    {
      "name": "Library strategy",
      "concept_code": "Biomaterial.library_strategy"
    },
    {
      "name": "Analysis type",
      "concept_code": "Biomaterial.analysis_type"
    }
  ]
}

Python versions

This package supports Python versions 3.7 - 3.10.

Package management and dependencies

This project uses pip for installing dependencies and package management.

• Dependencies should be added to requirements.txt.

Testing and code coverage

• Tests are in the tests folder.

• The tests folder contains tests for each of the tools and a test that checks whether your code conforms to the Python style guide (PEP 8) (file: test_lint.py)

• The testing framework used is PyTest

• Tests can be run with python setup.py test

Coding style conventions and code quality

• Check your code style with prospector

• You may need run pip install .[dev] first, to install the required dependencies

License

Copyright (c) 2019 The Hyve B.V.

The CSR to TranSMART loader is licensed under the MIT License. See the file LICENSE.