Genetic data normalization for genomic medicine: a Fast Healthcare Interoperability Resources Genomics reference implementation

Abstract

Objectives Demonstrate the ability to encapsulate clinical-grade genomics data normalization algorithms within a FHIR Genomics reference implementation. Background Variability in genomics data representation is a significant impediment to precise search, clinical decision support rule writing, variant annotation, and more. Such variability is problematic not just for genetic variants, but also applies to HLA alleles, phenotype codes, and more. Here, we provide an overview of genomic data variability and normalization algorithms, focusing on three key areas: genetic variants, HLA alleles, condition and medication variant annotations. We describe and demonstrate the strategies used in a public open source FHIR Genomics reference implementation. Materials and Methods We developed a set of design considerations, which we used to weigh different normalization approaches. All data (ingested patient data, ingested knowledge, query parameters) are subjected to normalization. Variant normalization leverages the biocommons/hgvs python package. HLA allele normalization leverages the py-ard python package. For variant annotation terminology variability (for conditions and medications), we leveraged FHIR-based ConceptMaps. Results Algorithms for normalization of genetic variants and HLA alleles, and terminology translations, have been implemented and deployed in a public open source FHIR Genomics Operations reference implementation. All data and source code described in this report are located at https://github.com/FHIR/genomics-operations, and deployed at https://fhir-gen-ops.herokuapp.com/. Every normalization strategy examined to date has known limitations. Conclusion While we report on our experience successfully encapsulating genomic data normalization in FHIR Genomics Operations, the challenges and solutions identified are broadly applicable to many other contexts.