Next generation sequencing is considered a diagnostic technique, and, over the past decade, molecular assays based on this technology have been integrated into the service portfolio of the public health care system in Spain.

Currently, the implementation of NGS methods is uneven across autonomous communities and partly determined by the technological resources and staff trained in bioinformatics required to ensure the validity of the findings.

Their introduction in molecular laboratories has marked a paradigm shift: in the past, technicians performed the experiments and clinicians issued the report. Now, NGS requires data analysis prior to clinical interpretation, which in turns requires that laboratories adapt to meet these new demands.

The main distinctive features of NGS as a diagnostic technique can be summarized as (1) decentralization and (2) the lack of built-in quality control checkpoints.

Decentralization involves the execution of multiple sequential steps to obtain a clinically interpretable result that may be performed by different laboratories or institutions. Each step requires specific facilities, specialized resources and qualified staff.

This workflow (Fig. 1) begins with the physician ordering the test (after obtaining written informed consent), continues with DNA extraction from the tissue of interest and the creation of DNA fragment libraries using standardized protocols, proceeds to DNA sequencing on a high‑throughput sequencing platform and culminates with a bioinformatics analysis.

Figure 1.

Flow chart of the NGS workflow.

The stages involved in performing an NGS test are outlined below: first, the patient visits the facility and the requesting physician (specialist or clinical geneticist) orders an NGS test after obtaining signed informed consent (stage 1). The test is then carried out sequentially (stage 2) so that, finally, the results in the NGS report can be communicated to the patient in the office and, if indicated, genetic counseling can be offered to other family members (phase 3).

Subsequently, a clinical specialist prepares an NGS report that includes the findings of the test (positive, negative or inconclusive) and other relevant information about it.

Decentralization allows the entire test or some of its phases to be outsourced, giving rise to the following implementation models: (a) in-house, (b) outsourced, or (c) hybrid.

The fragmentation of the workflow into subprocesses requires standardized communication protocols between all parties to ensure traceability and correct management of the diagnostic process. Process oversight and monitoring are essential to ensure analytical and clinical validity,8 even in outsourced or hybrid implementation models.

The overall validity of the test largely depends on the analytical process, as it is during the bioinformatics analysis that the majority of issues (such as sample swaps, contamination or low-quality samples) are identified, which ultimately helps determine which samples are suitable for diagnosis.

However, one of the challenges posed by NGS is the lack of built-in quality control checkpoints beyond those present in the sequencing platforms. This is of critical importance, especially in variant calling, where poorly calibrated pipelines or incomplete raw sequencing output can generate files with deficient information.

Furthermore, the lack of standardized best practices and decentralization (in cases where several laboratories are involved in performing the test) complicate the implementation of corrective measures that ensure quality and process integrity. This underscores the importance of laboratories adopting quality frameworks, such as ISO standards, to ensure patient safety. We will address these issues later in the text.

Another important aspect of NGS is the ability to perform targeted sequencing using disease-specific gene panels, focusing on specific genomic regions to reduce costs and analysis time. Similarly, when broader analysis is required, whole exome sequencing (WES) and whole genome sequencing (WGS) are possible alternatives.

Last of all, we must highlight the need to manage the large data volumes generated by each test. The generated file size varies according to multiple factors, such as the size of the sequenced region, the depth of coverage (also referred to as read depth), and the number of samples sequenced simultaneously.

For instance, the specifications for the Illumina NovaSeq 6000 system9 provide a general idea of the magnitude of the files that are generated in these processes: run output folders range from approximately 230 gigabytes (GB) to 3 terabytes (TB), whereas a complete test, considering all the files generated during both sequencing and analysis, may add up to between 560 GB and 6.5 TB of storage.

NGS data must be treated as medical data, just like medical imaging data or any other clinical information. From this perspective, and in accordance with Law 41/2002 on Patient Autonomy, they must be retained for a minimum of 5 years.

Therefore, having an adequate storage infrastructure is essential, along with efficient planning to enable medium‑term retention of genomic data.

In addition to genomics, other omics technologies are gaining relevance in clinical practice. Although they are already sometimes used for diagnostic purposes, their application remains largely confined to preclinical research.

These technologies play a key role in the discovery, analysis and validation of new therapies, biomarkers and diagnostic tools, which are fundamental to the advancement of personalized medicine.

In recent years, there have been significant advances in proteomics, which analyzes the presence of proteins in a sample on a large scale, thanks to technological improvements and more effective bioinformatics analysis.10 One salient example is the Overa test, developed by Vermillion, which uses tandem mass spectrometry (LC-MS/MS) to assess the risk of ovarian cancer in women with pelvic masses. This test was approved by the United States Food and Drug Administration (FDA) in 2016.

The analysis of metabolites is already an established practice in pediatric care, especially in the context of newborn screening for congenital metabolic disorders. Metabolomics constitutes a significant advance by enabling the untargeted analysis of thousands of metabolites, facilitating the identification of metabolic signatures in fluids and tissues. Research is currently underway to apply metabolomics to diseases in which early detection could improve patient outcomes, such as type 2 diabetes.11

Another field of interest is precision oncology, where identifying biomarkers is essential to enable early detection, tumor classification and treatment response monitoring. In this regard, epigenomic analysis has provided notable success stories, such as the use of EGFR‑activating mutations as biomarkers, which has significantly improved survival in patients with non‑small-cell lung cancer (NSCLC).12

In the field of molecular genetics, it is worth noting the advent of multiomics integration methodologies, which enable integration of information to improve the diagnosis of patients who exhibit characteristic phenotypes but in whom the causal disease mechanism has not yet been identified.

Thus, the inclusion of RNA sequencing (RNA‑seq) data in molecular studies allows analysis of the patient’s transcriptome, facilitating the identification of aberrant splicing and gene expression.13

Epigenetic signature analysis14 is also proving useful in patients with neurodevelopmental disorders, as it allows classification of patients with complex syndromic presentations or differentiation of overlapping phenotypes, such as Kabuki syndrome and CHARGE syndrome, thereby improving diagnostic precision.

Fig. 2B presents the typical workflow of a pipeline. There is usually an initial stage that includes (1) processing the sequencing output, (2) filtering the data, and (3) aligning DNA reads to a reference genome.

On Illumina sequencing platforms, the primary output from the sequencer must undergo a demultiplexing step. The purpose of this process is to separate the raw data generated by the instrument into individual samples and assign each sample its corresponding FASTQ files, which contain the DNA sequence reads (hereafter, “reads”). When sequencing is performed in paired‑end mode, reads are obtained from both ends of each DNA fragment, which results in two FASTQ files per sample.

Next, a trimming step is applied to remove low‑quality reads, sequencing artifacts, and adapter sequences. Each trimmed read is then aligned to a reference genome, and the choice of reference genome can affect variant calling,15 so it is an important aspect to consider.

Aligned reads are stored in BAM (binary alignment/map) files, which can be opened and viewed in genome browsers. Manual review of regions of interest in these viewers provides experts with additional, context‑rich information that complements automated analyses.

Each read has an associated quality score that depends on multiple factors. For example, in short‑read sequencing, low‑complexity or repetitive genomic regions are typically covered by lower‑quality alignments, which can lead to coverage bias and hinder variant detection.

After mapping each DNA read to the reference genome, genetic variants are identified in a process known as variant calling. This step is highly specialized, because the algorithms used differ based on the type of variant that is to be detected.

These algorithms are based on statistical methods and require an adequate read depth to identify patterns. Fig. 2A shows a single nucleotide variant (SNV) in the AGL gene.

Figure 2.

Summary of the bioinformatics analysis process.

(A) Example of a heterozygous SNV in the AGL gene. Aligned reads are shown in grey and the nucleotide sequence of the featured exon fragment can be seen at the bottom. (B) Sequential steps performed in the bioinformatics analysis process that must be implemented through a pipeline. (C) Information for the variant shown in (A) with the corresponding annotations provided to support the interpretation of the variant.

In addition to identifying the variant, it is essential to determine whether it is heterozygous or homozygous (that is, whether the variant is present in one or both alleles). However, the sequencing workflow generally does not preserve information about the allelic origin of each read, so this must be inferred later in the process.

Variant calls are subject to uncertainty of varying degree, which may give rise to false positives. Therefore, the reliability of the findings must be evaluated using the quality scores provided by the available tools, taking into account biases that may affect specific genomic regions. In addition, viewing variants in their genomic context is recommended for a more accurate interpretation.

SNV calling algorithms have evolved over the past decade, and samples are now analyzed in groups rather than individually to minimize the impact of recurrent artifacts within sequencing batches (joint calling).

GATK416 is the best-known toolkit for the purpose, while DeepVariant,17 which is based on neural networks, can be applied in different contexts, including somatic variant calling. An article published by Barbitoff et al. (2022)18 offers an overview of bioinformatics tools for calling SNVs and INDELs.

To detect CNVs, the most common method estimates gene dosage from normalized read depth relative to a control population, an approach frequently used in targeted sequencing assays. In whole‑genome sequencing (WGS), multiple strategies are combined to identify CNVs and structural variants such as large deletions, duplications, inversions and translocations.19

Other types of variants that are of interest, especially in hereditary neurodegenerative diseases, are tandem repeats and trinucleotide repeats. To learn more about their detection by NGS, we recommend the review published by the Genomics England Research Consortium.20

Unlike SNVs/INDELs, the detection of SVs using short-read platforms is mainly limited to screening due to the high sensitivity and low specificity of the algorithms, which result in a high false-positive rate. This limitation arises from technological constraints and designs limited to exon sequencing to maximize cost-effectiveness. In clinical practice, hybrid approaches combining multiple technologies are currently being explored following promising results in research.

It is recommended that CNVs and SVs detected by short-read sequencing be confirmed through an orthogonal method. However, this is not always feasible due to the lack of commercial MLPA kits (MRC Holland) or CGH/SNPs arrays covering the regions of interest. Therefore, it is crucial that the reportable range is defined in the test development phase, excluding regions for which there is no clear validation method.

In this step, additional functional, population and clinical information (a process known as annotation) is provided for the identified variants to support their interpretation.

The standard HGVS nomenclature21 should be used to describe SNVs and INDELs based on a carefully selected transcript.22,23 Variant annotation can change based on the selected transcript or transcripts and could affect the accuracy of diagnosis.

Functional data has to be supported with in silico predictions using software applications designed to estimate the probability that an SNV or INDEL has a deleterious impact.24Table 5 lists the most widely used predictors.

Population-scale aggregated data helps identify variants in the reference population. The gnomAD database25 is one of the most comprehensive resources, as it has data for 195 000 individuals that serve as a reference population.

Annotation should also include information on known expert-curated, clinically important variants, available through databases such as dbSNP, ClinVar and HGMD (the latter requires a license for use).

Although the workflow for SNVs and INDELs has been standardized, tools for CNV and SV annotation remain limited and lack widely accepted standards. The use of tools such as ClassifyCNV and AnnotSV, as well as CharGer, which incorporates criteria from the American College of Medical Genetics and Genomics and Association of Molecular Pathology (ACMG/AMP),26 considered the standard framework for variant classification, should be considered.

Since there is no clear definition of the role and scope of practice of clinical bioinformaticians, there are differing views on who should be responsible for variant filtering and prioritization.

This step should preferably be performed by experts in molecular genetics, who would be responsible for determining the pathogenicity of candidate variants according to the ACMG/AMP guidelines.

However, more complex analyses require the use of virtual gene panels and/or advanced filtering with specialized or custom-made algorithms. Therefore, close collaboration between bioinformaticians and molecular geneticists is essential to support decision-making in the final phase of clinical NGS.

Various filtering and prioritization tools are available to facilitate these tasks; most require a license, allowing handling of extensive lists of variants, although their description is beyond the scope of this review.

Analysis methods are constantly evolving, so it is essential to upgrade pipelines with more effective algorithms and updated databases. Since there are no regulations governing the frequency of these updates, it is crucial to document pipeline versions including detailed descriptions of the software and resources used, and to include this information in NGS reports to trace the analysis process.

Pipelines need to be monitored to assess their performance over time and ensure their validity when changes are introduced in the diagnostic process. Also, in the case of in-house pipelines, it is essential that program parameters are configured correctly in the development phase.

Therefore, well-characterized reference samples, such as those provided by the Genome in a Bottle (GIAB) consortium,27 need to be used to optimize performance and assess the reproducibility of the results. However, reference data are not available for some alterations, which limits the implementation of well‑calibrated algorithms in routine diagnostic practice.

For instance, in the case of the GIAB NA12878/HG001 reference genome, vials with DNA samples of this particular individual can be obtained to validate the bioinformatics process alone or the entire workflow from the initial experimental stages.

Regular evaluations can also be performed to assess performance against a standard or independent review, for instance, using EMQN external quality assessment schemes. Implementation of evaluations by external agencies is recommended to assess the performance of pipelines.

The reproducibility of the results obtained with the pipelines may be affected by the hardware and the software versions of the computational tools used in the analysis. For this reason, running pipelines in containers is recommended,28 as containers package the code along with all necessary dependencies and archives, guaranteeing consistency across environments and improving portability, traceability and reproducibility.