2.1 Legal and normative requirements for quality assurance

With regard to legal and normative requirements, quality assurance of genome diagnostics does not differ from that required for other human genetic diagnostic methods. However, due to the volume of data generated, the aforementioned potential for additional or incidental findings, and the possible use of equipment and software solutions outside the direct control of the facility, additional regulations within the legal and normative framework are required.

In Germany, the requirements of the Genetic Diagnostics Act (GenDG; http://www.gesetze-im-internet.de/gendg/index.html) apply. In addition, the implementation of diagnostic procedures must be based on RiLiBÄK. Though accreditation according to ISO 15189 is not mandatory for genetic diagnostics in Germany, this international norm provides an additional framework. In 2022, a revision of ISO 15189 was put in force, which must be implemented from 2025 onwards, following a period of transition. Other standards that are applied by laboratories, depending on the focus of their activities, are DIN EN ISO 17020 (applicable to the field of pathology/neuropathology) and DIN EN ISO 17025 (testing and calibration laboratories working in a non-medical context). In terms of software development, IEC 62304 (for “medical device software”) can provide assistance with respect to the requirements for, and the development of, software in genetic diagnostics. In addition to these laws and standards, which largely concern genetic analyses, since 2022, Regulation (EU) 2017/746 on in-vitro diagnostics has also been mandatory.

2.2 Objective of genome analyses

The purpose of diagnostic genome analyses is to determine both hereditary and somatic causes of disease. This means that the analysis of different primary samples, such as blood and tissue, may be necessary, depending on the indication. In current diagnostics, “short-read” sequencing techniques with read lengths of 100–300 bp are mainly used. Depending on the characteristics of the gene of interest, these can be complemented by “long-read” sequencing with read lengths of 10–100 kb (14).

In a number of syndromic diseases, pathogenic variants occur regularly in a mosaic state (e. g., Proteus syndrome (AKT1), tuberous sclerosis (TSC1)). In diseases such as neurofibromatosis type 1, while variants in the affected tissue may be mosaic, they may not be detectable in blood. In addition, the detection of somatic variants is used as a biomarker of treatment response in oncology patients, e. g., in PARP inhibitor therapy (15). Therefore, bioinformatic pipelines and quality parameters with an adequate limit of detection (LoD) should be used (6). A comprehensive human genetic evaluation may also include the detection of variants in the mitochondrial genome. The particular characteristics of these variants, such as tissue differences in “mosaic status” (16, 17) should be taken into account. In addition to determining sensitivity and precision, validation also should include determining the LoD for the detection of heteroplasmic mitochondrial variants. The presence of NUMTs (nuclear mitochondrial DNA sequences, “mitochondrial pseudogenes”) in the nuclear genome can lead to artifacts during the detection and interpretation of variants.

In addition to the recognition of SNVs, CNVs, and SVs, bioinformatic tools also enable the detection of “runs of homozygosity (ROH)”, which can indicate the presence of uniparental (iso-)disomy (UPD), and thus an autosomal recessive or imprinting disease. In trio analyses, UPD can also be determined without ROH detection. Secondary, somatic events in the sense of clonal hematopoiesis, which lead to a copy-neutral loss of heterozygosity and may also mask pathogenic changes, can also be captured via ROH analyses (18).

In addition to the diagnostic investigation of monogenic-, oligogenic-, and (tumor) predispositions, variant detection methods are also used for the assessment of multifactorial diseases and pharmacogenetic factors. The calculation of polygenic risk scores to determine relative risks on the basis of association studies has now been introduced into human genetic diagnostics.

When performing prenatal analyses, the exclusion of maternal contamination continues to be mandatory. During a prenatal trio diagnostic investigation, this can be achieved directly from the NGS data. In the context of non-invasive prenatal diagnostics (NIPT), the analysis of cell-free fetal DNA in the maternal plasma in order to detect aneuploidy and other genomic changes must be conducted in accordance with specific quality requirements (19).

3.1 Personnel

All personnel involved in genome diagnostics must be suitably qualified for their respective role (pre-analytics, sample collection, wet laboratory, bioinformatics, diagnostics, validation, medical evaluation), and have the appropriate training, experience, and skills (see ISO 15189:2012: 5.1). In this context, besides the legal requirements of the GenDG, the S2k guideline of the German Society for Human Genetics (GfH) (20) must also be considered. In addition to adequate staffing for the responsible supervision of technical processes, data interpretation, and reporting, this specifies the need for the integration of appropriately trained scientists (e. g., human geneticists) and physicians. An introduction into the specific working roles and ongoing training must be carried out and documented. In addition to professional qualifications (see above), individuals responsible for data evaluation and reporting must have several years of diagnostic experience in the molecular genetic analysis of sequence changes for genetically heterogeneous diseases (including raw data and variant evaluation).

3.2 Premises

The premises must fulfil the requirements of modern NGS diagnostics in terms of equipment, space, layout, and house-keeping (electrical system, air conditioning, statics, access-control). Working steps that are at risk of cross-contamination must be physically separated. Accordingly, the separate storage of primary samples, nucleic acids, control materials, reagents, consumables, and analysis products must be observed. In the area of data processing, all hardware components must be kept in a location that ensures the physical security of the data.

3.3 Consumables, equipment and software

The laboratory must have a documented procedure for the procurement and handling of equipment, including maintenance, servicing, storage, and disposal. Batches of consumables must be recorded and managed in accordance with normal standards. For all equipment, (functional) monitoring must be documented in accordance with ISO 15189. This applies to all equipment used in the laboratory processes, from small devices (e. g., pipettes) to thermal cyclers and measuring devices for quality and quantity determination of nucleic acids to the NGS platforms and associated IT systems. On implementation, all components and modules used in the procedures must also be considered in the validation process, and their validation must be documented. When appropriate, this documentation should be performed in a modular manner. Significant changes, including updates/revisions of the chemistry, the sequencing devices, or the evaluation software, must be reverified or revalidated.

Due to the high investment and maintenance costs, NGS platforms and evaluation software are sometimes used jointly by different institutions, while external providers, including commercial ones, may be used for analysis software and data storage. In principle, this cooperation is possible. However, the relevant provisions (General Data Protection Regulation, GenDG, ISO 15189) must be taken into account. For example, shared use of a device must be documented and regulated in such a way that the genetic testing facility retains complete control over the process. This includes restrictions of access to unauthorized person in order to protect the device and the correctness and security of the patient data and samples. In addition to these security measures for data and sample protection, the use of cloud-based evaluation tools must take into account the laboratory’s need to access the data at any time, and that availability and storage must be possible over the period for which documentation is required by law.

3.4 Preanalytics

In addition to the requirements specified in ISO 15189 (instructions for sample collection and storage), the prerequisite pre-analytical measures for NGS analyses must include consent for sample collection, implementation of the investigation, and the processing of the genetic data, as stated in § 8 of GenDG. Here, the requirements for the informed consent procedure that are specified by GEKO, including the qualification requirements for the responsible physician, must be taken into account (10).

As formulated in the ESHG guidelines for NGS diagnostics (2) and in the S1 guidelines of the GfH (5), the laboratory is tasked with providing the referring clinician with a publicly accessible list of disease-relevant genes (“target genes” in the panel offered), which defines the scope of the investigation. Information must also be provided concerning the type of genetic variant (SNVs, CNVs, SV, repeats) for which the test is validated, and the limitations of NGS diagnostics (EuroGen recom. 5, 6 (6)).

3.5 Implementation of NGS procedures

The following sections present the various steps of the diagnostic NGS workflow, with special emphasis on quality assurance measures (Fig. 1).

Figure 1:

Formal levels of quality requirements in the Next Generation Sequencing process

3.5.2 Data processing

An NGS bioinformatic pipeline includes algorithms that process sequencing data and associated metadata, from the raw data to variant annotation and variant prioritization. This process may involve multiple software components, databases, and operating environments (hardware and operating system). In addition, diverse file types are used and generated during data analysis (Tab. 1, Fig. 2). Standard data formats, such as FASTQ, BED, SAM, BAM, CRAM, and VCF, are widely used and recommended (EuroGen recom. 17).

Table 1:

Short-read next generation sequencing file formats. (*BAM and CRAM files also contain non-aligned sequences).

File type	Content
BCL	binary base call—Contains the raw sequence data
FASTQ	Sequenced reads with corresponding quality values.
BED	Browser Extensible Data: Generic format for specifying genomic regions as a combination of genomic coordinates and annotations.
SAM	Sequence Alignment Map: Information concerning reads and their mapping in the genome.
BAM*	Binary Alignment Map: Compressed binary version of the SAM file.
CRAM*	Compressed Reference-oriented Alignment Map: Highly compressed version of the SAM file.
VCF	Variant Call Format: Generic format for describing genetic variants.
GVCF	Genomic VCF format with information related to each base in the genome irrespective of the presence of a variant

Figure 2:

Data storage requirements in NGS diagnostics using the example of exome and genome sequencing. Data flow and exemplary ES and GS dataset sizes are shown. (* Average file size for 1 sample at 125x coverage and a read length of 2x150 bases. File size may vary slightly, depending on enrichment, compression level, and which instrument platform is used (due to different base quality score encoding). ** Average size for VCF files generated using the “GATK best practice workflow”. This can vary substantially, depending on the applied variant caller, filter strategy, and compression level. ***Average file size for 1 sample at 40x coverage and a read length of 2x150 bases. File size may vary slightly, depending on the compression level and the applied sequencer (due to different base quality score encoding). ^$Conversion of BAM/CRAM format back to FASTQ leads a loss of information, since only the first part of the Illumina read name is overtaken in the BAM/CRAM file format. Information on reads 1 or 2 in paired-end sequencing, pass filter, and the index sequence are lost.

The workflow for clinical NGS-based data processing comprises three main process steps: a primary analysis for generating DNA sequence data from the raw data; a secondary analysis, which includes read alignment to a reference genome and variant calling; and a tertiary analysis for variant annotation, filtering, interpretation, and reporting (7). Although the bioinformatic process for the analysis of GS does not differ substantially from that used for ES, additional algorithms are often necessary, e. g., for SV detection, repeat expansions, and annotation of non-coding regions.

3.5.2.1 Primary analysis

The primary analysis includes base calling, i. e., the generation of sequence- and quality data (PHRED scores), and demultiplexing of the raw data, which are usually saved in FASTQ format. The PHRED score is an important measure for the correctness of an individual base call, and can be influenced, for example, by technical artifacts. Here, the “GATK best practice workflow”, among others, enables the partial correction of these errors by recalibrating the basic quality score (BQSR) via machine learning. Artifacts in NGS alignments are device-independent, and often result in de novo false positive variants. During variant filtering, systematic errors can be reduced via a comparison with data sets generated on the same sequencer. Removing adapters and trimming poor-quality bases ensures that only high-quality, optimal-length reads are used for downstream analysis.

3.5.2.2 Secondary analysis

The secondary analysis is the computationally most intensive aspect of an NGS analysis. The choice of both the reference genome (GRCh37/hg19, GRCh38/hg38, T2T-CHM13) and the alignment algorithm (e. g., for DNA BWA-MEM, DRAGMAP) can have a major impact on the accuracy of complex variant calls (21). While improved variant detection and genotyping have been demonstrated for the recently published T2T-CHM13 assembly (22), its use in clinical diagnostics has not yet been adequately evaluated. A broad diagnostic application is currently precluded by the lack of transference for variant databases (e. g., ClinVar, gnomAD) to the genomic positions in the T2T-CHM13 genome. Benchmarking studies show that the use of GRCh38/hg38 improves variant detection, and generates fewer false-positive variants, compared to the use of GRCh37/hg19, and its use is therefore recommended by Eurogentest (EuroGen recom. 15 (6)). However, some databases only contain datasets for either GRCh37 or GRCh38, and in that case, the “complete” annotation of variants is only possible via a data liftover. Due to erroneously integrated duplications in the GRCh38 assembly, variant detection for several genes (e. g., CBS, KCNE1, and CRYAA) is worse compared to GRCh37. However, this problem can be solved by masking these areas in the reference genome (21). One approach to improving variant detection in these genes and in other genes with homologous copies in the genome (e. g., SMN1, PKD1, NEB, GBA1, STRC) is to use sequence graphs in the alignment algorithms (23).

Aligned sequences are stored in compressed BAM and CRAM files, together with information on their position in the genome (Tab. 1, Fig. 2). To avoid artifacts and false-positive variants, duplicates should be marked or removed, and local realignment performed. In contrast to ES, library preparations in GS are generated PCR-free when sufficient DNA is available. This reduces duplicate reads to 3–5 %, compared to 5–15 % in a typical ES procedure (24). The capture of GC-rich areas, on the other hand, is superior, which leads to a uniform sequencing depth and diagnostic coverage.

As described, GS opens up the possibility of recognizing complex variant types that can only be captured to a limited extent with ES (SVs, REs). It has been suggested that in order to maximize the sensitivity of detection, pipelines should use several variant detection tools (Table 2), i. e., tools that are specific to each variant class (24). Combining the results of two orthogonal SNV/InDel callers (e. g., GATK HaplotypeCaller, DeepVariant, Strelka2), and then merging multiple variant callsets (in VCF format) into a single callset using software packages such as BCFtools can provide a slight sensitivity advantage. However, care should be taken to ensure correct handing of complex variants and/or differences in variant representation (8).

Table 2:

Frequently used software modules for variant calling, quality checking, and other processing steps necessary in short-read sequencing. (Information on the named software is available upon request.)

Analysis step	Processing step	Software (example)
Primary analysis	Demultiplexing	bcl2fastq
	Alignment	BWA-Mem, DRAGMAP
	File management	BCFtools
	SNV calling	GATK HaplotypeCaller, DeepVariant, Strelka2
Secondary analysis	CNV calling	MOPS, CNVkit, CONTRA, CoNVEX, ExomeCNV, ExomeDepth, GATK, XHMM
	Structural Variant calling	DELLY, Lumpy, Manta, Pindel, SVMerge
	Repeat expansions	ExpansionHunter, exSTRa, STRetch, TREDPARSE
	TEs and insertions	Mobster, MELT, TraFiC-mem
	Low frequency variants (mosaic and somatic)	MuTect2

The laboratory should validate both the impact of the aligner and the accuracy of variant detection for all detectable variant classes, i. e., for SNVs, InDels, CNVs, and complex genomic structural variants (SVs, REs) separately (EuroGen recom. 22 (6)). Sequencing of the whole genome represents the most comprehensive approach, which typically yields an average sequence depth of around 30–60x across the entire genome. However, the higher sequence depth achieved in panel- and exome sequencing may allow a more sensitive detection of low allelic frequency variants, e. g., subclonal somatic mutations in tumors or variants that present as germline mosaics (24). The reliability and accuracy of algorithms for the detection of SVs and REs from GS data remain suboptimal at present. However, they allow the detection of a large number of pathogenic changes, and currently tend to be used as a screening method. Before reporting such complex variants, technical confirmation may be required using an alternative, diagnostically established method.

3.5.2.3 Tertiary analysis

The first step in tertiary analysis is variant annotation, which involves the naming of each variant using a standardized nomenclature, followed by the linking it to information from various databases, the medical literature, and information concerning the quality of variant detection (e. g., the number of reads in which the variant occurs). Here, the nomenclature guidelines for SNVs from the Human Genome Variation Society (HGVS), and the nomenclature guidelines for CNVs from the International Standing Committee on Human Cytogenetic Nomenclature (ISCN), should be used (Table 3). Correct annotation is required for the subsequent filtering, prioritization, classification, and interpretation of variants (see 3.6). For complex genomic variants, such as multinucleotide variants (MNVs)—which are usually listed independently, but which must be considered jointly in order to allow correct assessment of their pathogenicity—annotation remains challenging. The correct detection and annotation of complex structural variations, such as inversions and balanced translocations, can also be difficult using GS data alone, and may require the use of an additional method.

Table 3:

Further information on selected nomenclature guidelines and reference materials.

Name	Website
Human Genome Variation Society (HGVS)	https://varnomen.hgvs.org/
Human Gene Nomenclature (HUGO)	https://www.genenames.org/
Matched Annotation from the NCBI and EMBL-EBI (MANE)	https://www.ncbi.nlm.nih.gov/refseq/MANE/
International Society of Cytogenetic Nomenclature (ISCN)	https://iscn.karger.com/
Genome-in-a-bottle	https://www.nist.gov/programs-projects/genome-bottle

EuroGen recommends that all GS variants identified at a given site should be annotated (EuroGen recom. 23), and that the aggregated data should be compiled in order to generate internal allele frequencies and allow the interpreted data to be shared with other institutions (EuroGen recom. 24 (6)).

3.5.2.4 Pipeline quality control and validation cycles

The bioinformatics pipeline should have sufficient quality control measures to ensure that the generated data are robust, and that error detection and compliance are provided (EuroGen recom. 16, 21 (6)). To enable quality control of the bioinformatics process, quality data can be used in each step of the analysis (Table 4). The NGS pipeline should be platform-specific and tailored to the needs of the laboratory (EuroGen recom. 17 (6)).

Table 4:

Relevant quality parameters in the NGS process that are also recorded within the framework of the EMQN Quality assessment schemes.

Parameter	Definition
Yield	Throughput of the sequencing run.
Error Rate	Percentage of mismatched bases per cycle, as determined by alignment to PhiX.
%Q30	Percentage of bases identified with a quality score of 30 (equivalent to 1 error in 1000 bases, Phred score) or above.
Density (K/mm2)	Cluster density in thousands per mm2.
Cluster PF (%)	Cluster densities that satisfy the internal chastity filter and can therefore be used for downstream analysis.
Phas/Prephas (%)	Percentage of molecules in a cluster that gets out of sync in each cycle due to incorporation of multiple nucleotides (pre-phasing) or due to failed incorporation of a nucleotide (phasing).
Percentage mapped reads	Percentage of reads that could be mapped to the reference sequence.
Duplicates (%)	Percentage of duplicates not typically used for downstream analysis.
Insert Size	Average size of sequenced DNA molecules.
GC content (%)	Average GC content of the sequenced DNA fragments.
Mean Coverage	Average number of reads covering a specific target region.
Cov30	The average number of bases sequenced that match or “cover” known reference bases. For example, an entire genome sequenced with 30-fold coverage means that each base in the genome has been sequenced an average of 30 times.
Uniformity	Coverage uniformity describes the distribution of reads across the target regions in the genome.

Bioinformatic pipelines developed for the analysis of GS/ES clinical data are complex and require a robust quality assurance program for ongoing monitoring and pipeline updates. Validations must document all steps required to compile, install, and run specific GS clinical bioinformatic pipelines. Both a version control system and the detailed documentation of code changes are required for the deployment of all software updates ((25); IEC62304).

Pipelines should be tested against reference standards (e. g., genome-in-a-bottle samples) to ensure they are operating in a reproducible and error-free manner (26). Validation is performed on a specific version of the software, and the data analysis should be repeated whenever software components or reference data files are updated. Laboratories should have a Release Management process, with a clear definition of “major” and “minor” changes that define the level of validation. If relevant changes are made to software components or parameters, the testing mechanism should revalidate both the local test performance and the impact of these changes on the clinical process of variant detection and annotation. Ideally, software validation should include continuous integration processes for updates and enhancements. For incremental changes, software testing may need to be automated to some extent. Crucially, errors must be tracked down effectively, and underlying issues must be identified and resolved.

3.6 Post-analytics

This issue is defined by ISO 15189, and stipulates, among others, the requirements for the release of test results. In addition to the aforementioned criteria for the evaluation of results, post-analytics should include the plausibility check, e. g., in order to allow the collection of additional information using alternative methods (e. g., Sanger sequencing, Multiplex Ligation-dependent Probe Amplification/MLPA) when the examination results are not sufficiently reliable in terms of the test request.

3.7 Reporting of findings

The structure and content of reports are based on the requirements of ISO 15189 and the S2k guideline for human genetic diagnostics (20). In addition to information on patient identification and the indication for the test, the first page should include a concise description of the result that is of relevance to the sender in terms of diagnosis, prediction, therapy, or genetic counseling (EuroGen recom. 38 (6)). As with other reports of findings from genetic analyses, the method must also be clearly described. This description should include all of the essential evaluation parameters, in particular information on the test procedure; the variant types that can be detected with the method; the coverage and reading depth of the target region (region of interest), i. e., the exome, genome, and/or the genes that are reported explicitly; as well as analytical sensitivity and precision (EuroGen recom. 37 (6)). The report should also include details of the reference genome, reference transcript version, and OMIM reference (EuroGen recom. 39, 40 (6)), as well as information on variant assessment (e. g., ACMG). The versions proposed by the MANE initiative should be used as reference transcripts. The extent to which database and literature references, as further explanations of the clinical relevance of (probably) pathogenic variants, are mentioned in the findings is to be defined by the laboratory. While all variants that are pathogenic or probably pathogenic according to the ACMG classification and that are of relevance to the indication for testing are reported, VUS should only be evaluated and reported if they are located in genes that are clearly associated with a known phenotype (EuroGen recom. 42 (2, 5, 6)).

3.8 Quality assurance measures

The laboratory must select, and then validate or verify, examination methods that are suitable for clarifying the given medical question. Guidelines and notes for validating NGS-based examination methods are available (2, 5, 6). For GS, special supplementary recommendations have been published (21). The validations of the platform and the pipeline should consider reference materials such as genome-in-a-bottle samples (GiaB, Tab. 4). The use of gene panels to clarify specific questions implies validation at the gene level. The coverage of known disease-causing variants should be ensured, and possible limitations of the methodology (e. g., pseudogenes) should be taken into account. Transcripts that have been harmonized across databases should be used as reference sequences (3).

In terms of sample tracking, genotyping should be established using an orthogonal method that is used to match SNVs (EuroGen recom. 29 (6)), if sample tracking is not secured in any other manner. As a minimum, quality control of the analyses should include monitoring of the coverage determined by the laboratory.

In the case of prenatal analyses, the exclusion of maternal contamination is a further mandatory quality assurance measure.

Quality assurance also includes participation in external quality assessment (EQA) schemes (proficiency testing, interlaboratory comparisons) which are able to assess whether or not the intended level of quality has been achieved. The requirements for this are specified in the new version of RiLiBÄK, which is currently being drafted. ISO15189 sets common specifications for quality assurance. The laboratory must select EQA schemes that cover all clinically relevant issues and which review the entire analysis procedure. This includes consideration of wet laboratory procedures, data processing, variant classification and interpretation, and reporting. Corresponding EQA schemes are offered by bodies such as BVDH, EMQN, UK NEQAS, and GenQA.