16. Converting VCF file to HL7 FHIR JSON Genomic Report profile

Recipe Overview
Reading Time
15 minutes
Executable Code
Yes
Difficulty
Expressing Clinical Genetic Information as FHIR JSON
FAIRPlus logo
Recipe Type
Hands-on
Maturity Level & Indicator
DSM-1-C0DSM-1-C1
hover me Tooltip text

16.1. Main Objectives

The main purpose of this recipe is to provide FAIR guidance relevant to the clinical domain by:

  • providing a tool to convert Variant Call Files (VCF) to a HL7 FHIR message

  • highlighting known limitations of the solution

  • raising awareness of the FHIR standard in the context of clinically relevant data.

  • discussing the benefits of obtaining genetic variation information in a regularized form and available in a well-known syntax.

16.2. Graphical Overview

16.3. User Stories

The table below lists relevant use cases.

As a ..

I want to ..

So that I can ..

Data owner

Convert VCF to a FHIR message

Produce an information payload carrying patient genotyping information in a standardized format compatible with EHR

Data consumer

Integrate patient genetic information

Have seamless integration with other FHIR messages from other sources

Data manager

Unify clinical information in one format

Facilitate reuse and mining by clinicians


16.4. FAIRification Objectives, Inputs and Outputs

Actions.Objectives.Tasks

Input

Output

Format conversion

FHIR format

Conversion results, Error report

16.5. Table of Data Standards

Data Formats

Terminologies

Models

VCF

BCF

FHIR

Compressed Format

LOINC


The Variant Call File or VCF is a file format specified by the Global Alliance for Genomic Health to report on genetic variation as detected by a range of molecular biology techniques (e.g., PCR, GeneChip, nucleic acid sequencing). It is considered to be the de facto standard for reporting genetic variations in their various forms. It is therefore the output for most genetic analysis pipelines (e.g., the Galaxy Worflow tool affy2vcf )

The latest version of Variant Call File format is v4.3, the detailed specifications of which can be found here

The VCF format is species agnostic, making it suitable for use in any context, from agronomy to clinical practice. In fact, it is this last use case that this particular recipe will be focusing on. Indeed, this is when bioinformatics meets medical informatics and the need to translate data into different format arises. In the world of clinical informatics, exchanging information between systems increasingly relies on Health Level 7 data standards and in particular on the Fast Healthcare Interoperability Resource (FHIR). A number of working groups focus on how to best fit clinical information within the paradigm of the HL7 FHIR representation.

In this FAIR Cookbook recipe, we will highlight a software component allowing to convert a specific type of genetic variation information stored in VCF files into an HL7 FHIR compliant, JSON formated message.

The aptly named vcf2fhir library is a python package designed to perform this task. It is the result of work recently published by Dolin et al, 2021. 1.

Warning

In its current form, vcf2fhir :

  • supports simple variants (SNVs, MNVs, Indels)

  • does not support structural variants

  • This software is not intended for use in production systems

  • This software is mainly a prototype for evaluation, testing and refinement of the conversion process and the FHIR message payload

  • This software does not deal with patient reidentification issues, which needs to be carefully considered if dealing with real patient data, and not test data.

16.5.1. Requirements

Users should be at ease with command line interfaces.

In order for the vcf2fhir python library to run, the following libraries need to be present on the system.

Software

Description

Version

cython

C-Extensions for Python

0.29.24

wheel

wheel

0.37.0

wget

File downloader

1.19.4

Users running a python environment (virtual or otherwise) need to install cython and wheel, which respectively provide C-implemented extensions for speed and a code packaging library which allows installations to run without the need for building and compiling. The ‘wget’ component will be used to obtain test files from a web address via a command line call.

Both python modules can be installed with a single invokation of the python installation pip command as follows:

pip install cython wheel

16.5.2. Installing vcf2fhir python library:

Since the vcf2fhir is available from pypi.org, we can also install the vcf2fhir binary using pip.

pip install vcf2fhir

16.5.3. Using vcf2fhir package:

In order to use the VCF to FHIR converter function provided by the library, one needs to first obtain VCF files. Not only that, but as we indicated in the introduction, the version of the VCF files should be at least v4.1. Furthermore, they should be such that they contain only simple variant information and not structural variants (a type of genetic variations which aren’t currently supported by the vcf2fhir library).

Note

  • The following section shows how to download VCF files available from public location, the VCF2FHIR github repository in this instance.

Obtaining an exemplar VCF file from the vcf2fhir github repository using the wget command:

wget -c https://raw.githubusercontent.com/elimuinformatics/vcf2fhir/master/vcf2fhir/test/vcf_example1.vcf
more vcf_example1

will show the following:

##fileformat=VCFv4.1
#CHROM  POS ID  REF ALT QUAL    FILTER  INFO    FORMAT  TEST001
1   15  .   .   G   .   .   .   GT:PS   0/1:.
1   16  .   A,T G   .   .   .   GT:PS   0/1:.
1   17  .   A   G   .   27  .   GT:PS   0/1:.
1   18  .   A   G   .   .   SVTYPE=INV  GT:PS   0/1:.
1   20  .       G   .   .   .   GT:PS   0/1:.
1   25  .   A   G   .   .   .   PS  .
1   26  .   A   G   .   .   .   GT:PS   ./.:.
1   27  .   A   G   .   .   .   GT:PS   .|.:.
1   28  .   A   <CGA_CNVWIN>    .   .   .   GT:PS   0/1:.
chr1    301 .   A   G   .   .   .   GT:PS   0/1:.
1   400 .   A   G   .   .   .   GT:PS   1/1:.
1   500 .   A   G   .   PASS    .   GT:PS   0|1:500
1   600 .   A   G   .   .   .   GT:PS   1|0:500
1   700 .   A   G   .   .   .   GT:PS   0|1:700

16.5.4. Conversion from VCF to FHIR

The conversion command can be issued as follows:

import vcf2fhir

vcf_fhir_converter = vcf2fhir.Converter('vcf_example1.vcf', 'GRCh37')
vcf_fhir_converter.convert()

The result of the conversion is a so-called FHIR Genomics report, the specifications of which are available here. A number of options are available from the converter to allow users to modify and tune the output to contains specific information depending on the use cases. The conversion can be restricted to a subset of records found in a VCF file by specifying particular portions, e.g., conversion regions, studied regions, clinical annotations, or uncallable regions.

For a full and detailed overview of these options, we direct the readers to the original manual for the vcf2fhir library.

Depending on the options specified by the user, different types of ‘FHIR genomics report’ may be generated. They will differ in content and layout but all rely on a number of normative patterns and terminologies (e.g. LOINC).

  • Create FHIR Diagnostic Report

  • Create RegionStudied observations

  • Create Variant observations

  • Create SequencePhaseRelationship observations

  • Create DiagnosticImplication observations

More examples to instantiate the converter

  • Converts all variants in VCF. FHIR report contains no region-studied observation.

vcf2fhir.Converter('vcftests.vcf','GRCh37', 'aabc')
  • Converts all variants in VCF. FHIR report assign homoplasmic vs. heteroplasmic based on:

    If allelic depth (FORMAT.AD)/ read depth (FORMAT.DP) is greater than 0.89 then allelic state is homoplasmic; otherwise, it is heteroplasmic.

    Note : the default value of ratio_ad_dp = 0.99 and the ratio_ad_dp is considered valid only when its value lies between 0 and 1.

vcf2fhir.Converter('vcftests.vcf','GRCh37', 'aabc', ratio_ad_dp = 0.89)
  • Submitting only noncallable region without other regions generates an error.

vcf2fhir.Converter('vcftests.vcf','GRCh37', 'babc', nocall_filename='WGS_b37_region_noncallable.bed')
  • Converts all variants in VCF. FHIR report contains one region-studied observation per studied chromosome.

vcf2fhir.Converter('vcftests.vcf','GRCh37', 'cabc', region_studied_filename='WGS_b37_region_studied.bed')
  • Converts all variants in VCF. FHIR report contains one region-studied observation per studied chromosome.

vcf2fhir.Converter('vcftests.vcf','GRCh37', 'dabc', region_studied_filename='WGS_b37_region_studied.bed', nocall_filename='WGS_b37_region_noncallable.bed')
  • Converts all variants in conversion region. FHIR report contains no region-studied observation.

vcf2fhir.Converter('vcftests.vcf','GRCh37', 'eabc', conv_region_filename='WGS_b37_convert_everything.bed')
  • Submitting only noncallable region without other regions generates an error.

vcf2fhir.Converter('vcftests.vcf','GRCh37', 'fabc', conv_region_filename='WGS_b37_convert_everything.bed', nocall_filename='WGS_b37_region_noncallable.bed')
  • Converts all variants in conversion region. FHIR report contains one region-studied observation per studied chromosome, intersected with conversion region.

vcf2fhir.Converter('vcftests.vcf','GRCh37', 'gabc', conv_region_filename='WGS_b37_convert_everything.bed', region_studied_filename='WGS_b37_region_studied.bed')
  • Converts all variants in conversion region. FHIR report contains one region-studied observation per studied chromosome, intersected with conversion region.

vcf2fhir.Converter('vcftests.vcf','GRCh37', 'habc', conv_region_filename='WGS_b37_convert_everything.bed', region_studied_filename='WGS_b37_region_studied.bed', nocall_filename='WGS_b37_region_noncallable.bed')
  • Conversion of a bgzipped VCF

vcf2fhir.Converter('vcf_example4.vcf.gz','GRCh37', 'kabc', has_tabix=True)

Below is a typical output of the vcf2fhir tool: a HL7 FHIR message compliant with the Genomics Report pattern. Note the use of LOINC terminology for key descriptors.

{
    "resourceType": "DiagnosticReport",
    "id": "",
    "meta": {
        "profile": [
            "http://hl7.org/fhir/uv/genomics-reporting/StructureDefinition/genomics-report"
        ]
    },
    "contained": [
        {
            "resourceType": "Observation",
            "id": "",
            "meta": {
                "profile": [
                    "http://hl7.org/fhir/uv/genomics-reporting/StructureDefinition/region-studied"
                ]
            },
            "status": "final",
            "category": [
                {
                    "coding": [
                        {
                            "system": "http://terminology.hl7.org/CodeSystem/observation-category",
                            "code": "laboratory"
                        }
                    ]
                }
            ],
            "code": {
                "coding": [
                    {
                        "system": "http://loinc.org",
                        "code": "53041-0",
                        "display": "DNA region of interest panel"
                    }
                ]
            },
            "subject": {
                "reference": "Patient/NB6TK328"
            },
            "component": [
                {
                    "code": {
                        "coding": [
                            {
                                "system": "http://loinc.org",
                                "code": "92822-6",
                                "display": "Genomic coord system"
                            }
                        ]
                    },
                    "valueCodeableConcept": {
                        "coding": [
                            {
                                "system": "http://loinc.org",
                                "code": "LA30102-0",
                                "display": "1-based character counting"
                            }
                        ]
                    }
                },
                {
                    "code": {
                        "coding": [
                            {
                                "system": "http://loinc.org",
                                "code": "48013-7",
                                "display": "Genomic reference sequence ID"
                            }
                        ]
                    },
                    "valueCodeableConcept": {
                        "coding": [
                            {
                                "system": "http://www.ncbi.nlm.nih.gov/nuccore",
                                "code": "NC_000001.11"
                            }
                        ]
                    }
                },
                {
                    "code": {
                        "coding": [
                            {
                                "system": "http://loinc.org",
                                "code": "51959-5",
                                "display": "Range(s) of DNA sequence examined"
                            }
                        ]
                    },
                    "valueRange": {
                        "high": {
                            "value": 201113567.0
                        },
                        "low": {
                            "value": 201038512.0
                        }
                    }
                },
                {
                    "code": {
                        "coding": [
                            {
                                "system": "http://loinc.org",
                                "code": "51959-5",
                                "display": "Range(s) of DNA sequence examined"
                            }
                        ]
                    },
                    "valueRange": {
                        "high": {
                            "value": 201378701.0
                        },
                        "low": {
                            "value": 201358008.0
                        }
                    }
                }
            ]
        }
    ]
}   

16.5.4.1. Built-in Support Conversion Error Logging

No conversion tool is failsafe. Therefore, the vcf2fhir library provides 2 distinct logging functions, which plug naturally into the python generic error logging package.

  • vcf2fhir.general: this mode provides the standard library logging functions.

  • vcf2fhir.invalidrecord: this mode logs all the records from the input vcf file which are in present in the conversion region but are not converted to fhir format.

To take advantage of this mechanism, users can invoke each of the vcf2fhir loggers in the manner described below:


import logging

# create logger
logger = logging.getLogger('vcf2fhir.invalidrecord')
logger.setLevel(logging.DEBUG)

# create console handler and set level to debug
ch = logging.FileHandler('invalidrecord.log')
ch.setLevel(logging.DEBUG)

# create formatter
formatter = logging.Formatter('%(asctime)s - %(name)s - %(levelname)s - %(message)s')
# add formatter to ch
ch.setFormatter(formatter)
# add ch to logger
logger.addHandler(ch)

16.5.5. Take home message from using the vcf2fhir python library

  • an initial capability supporting generation of HL7 FHIR Genomics Report message from VCF files.

  • generation of LOINC-annotated, JSON formatted documents containing simple genetic variation information.

  • availability of a conversion error log, for quality control and error tracking tasks.

16.6. Conclusion

Why does this matter and how does it relate to FAIR?

The conversion from VCF to HL7 FHIR JSON message has to do with the **I and R** of FAIR, that is interoperability and reusability.

  • From a syntactic standpoint, the availability of genetic variation information at a granular level in an easily parseable form (JSON) is a gain for anyone looking at merging this information with other clinical messages.

  • From a semantic standpoint, the reliance on LOINC vocabulary to mark up the patterns defined in the HL7 FHIR Genomics Reports enhances interoperation between systems by provided unambiguous annotations.

  • Finally, as more systems are able to produce FHIR messages from a variety of instruments or data sources, the availability of a FHIR message covering a subset of genetic variation available from testing facilities makes investigating and mining phenotypic and genotypic relations more straightforward.

  • However, one needs to remember that the capability afforded by the vcf2fhir library is at an early stage and only supports simple cases. More efforts and more efforts is needed before a functionality is available at a Technical Readiness Level compatible with production systems.

Any other important issues?

  • We have highlighted the existing limitations surrounding the use of the open source conversion tool and that users should carefully assess the nature of the information present in the input VCF files prior to executing the code. Bearing this in mind, the vcf2fhir tool provides an easy to deploy and easy to use solution for anyone interested in adding a FHIR message capability to a clinical genetic analysis pipeline,for instance on consuming DNA microarray GeneChip genotyping solutions. The authors of the tool aim to expand its capabilities to include enhancing the conversion logic to accommodate VCF rows representing structural variants (i.e. rows that contain an INFO.SVTYPE field).

  • Finally, it is important to realize that the resulting JSON message, as it is, lacks important metadata to be fully and properly FAIR (e.g., licence information). One has therefore to see this capability as one of the many elements that needs to be put together to build and deliver a FAIR infrastructure. For instance, this HL7 JSON message could be embedded in a more complex system, which would package information and deliver a FAIR payload.

16.7. References

16.8. Authors