Bioinformatic tools in the clinical diagnosis of genetic disorders in children: Focus on next-generation sequencing

INTRODUCTION

Genetic disorders contribute significantly to pediatric endocrine disorders, with many conditions arising from monogenic defects affecting hormone synthesis, binding proteins, transcription factors, ion channels, membrane receptors, and intracellular signal transduction pathways^[1] as well as from complex polygenic variation. To date, approximately 450 endocrine disorders have been identified in which genetic factors contribute substantially to disease pathogenesis.^[2]

Conventionally, the diagnosis of endocrine disorders has relied on clinical, biochemical, and imaging-based evaluation. However, with advances in molecular genetics and the increasing availability of next-generation sequencing (NGS) —including gene panels, whole-exome sequencing (WES), and whole-genome sequencing (WGS) —genetic analysis has become integral to the diagnostic workup, especially in cases with inconclusive findings or associated extra-endocrine manifestations.

While genetic test reports typically include variant classifications based on established guidelines, interpretations may vary across laboratories due to differences in classification approaches, available evidence, database usage, and interpreter-dependent factors. A study has reported discordance in variant pathogenicity classification in up to 46% of cases, with 37% of these discrepancies having potential clinical impact.^[3] Consequently, it is important for clinicians to have a basic understanding of variant interpretation, to critically evaluate reports, correlate findings with the patient’s phenotype, and, where appropriate, contribute to variant reclassification over time.

This review provides an overview of selected bioinformatic resources and tools useful to clinicians for interpreting genetic variants and understanding genomic reports for monogenic pediatric endocrine disorders, particularly those identified through NGS-based tests. Rather than providing an exhaustive technical discussion, this review focuses on practical tools that can assist clinicians in the routine interpretation of genetic test reports.

CASE SCENARIO

A 5-month-old male infant, first child of a nonconsanguineous couple with an unremarkable antenatal and perinatal history, presented with poor feeding, lethargy, and decreased activity following a minor illness. He had recurrent focal seizures with secondary generalization since 3 months of age (3–4 episodes/week), partially controlled with levetiracetam (20 mg/kg/day), along with global developmental delay. Social smile and maternal recognition were achieved at 4 months; he had visual fixation and tracking but no neck control or vocalization. Hearing was normal. There was no relevant family history.

On examination, he was drowsy, dehydrated, and tachypneic with acidotic breathing. His weight was 5.5 kg (–2.7 standard deviation [SD]) and length 62 cm (–1.9 SD). No dysmorphism was noted. Neurological examination showed generalized hypotonia, poor head control, and exaggerated deep tendon reflexes.

Investigations revealed hyperglycemia (random plasma glucose 580 mg/dL), hyponatremia (sodium 125 mmol/L), normal potassium levels (4.5 mmol/L), and high anion gap metabolic acidosis (pH 7.12, bicarbonate 9 mmol/L, partial pressure of carbon dioxide 20 mmHg, and anion gap 25) with elevated serum ketones, consistent with moderate diabetic ketoacidosis. Glycated hemoglobin was 12.2%, and fasting C-peptide was <0.1 ng/mL. Magnetic resonance imaging of the brain revealed mild cerebral atrophy, and electroencephalography showed hypsarrhythmia.

The child was managed with intravenous fluids and insulin infusion as per standard diabetic ketoacidosis protocol, followed by transition to subcutaneous basal —bolus insulin. WES identified a heterozygous missense variant in the KCNJ11 gene (NM_000525.4: c.602G>A; p. Arg201His), consistent with autosomal dominant developmental delay, epilepsy, and neonatal diabetes (DEND) syndrome (OMIM #618856). The evaluation of this variant and its implications for management are discussed subsequently.

DEFINITION OF BIOINFORMATICS

In the context of clinical genomics, bioinformatic tools refer to computational resources, software programs, and curated biological databases that enable the analysis, annotation, prioritization, and interpretation of genetic variants generated through high-throughput sequencing technologies such as NGS. These tools provide information on gene function, population variation, disease associations, and predicted pathogenicity, thereby assisting clinicians in the identification of genetic variants for diagnosis, prognosis, and personalized therapeutic strategies.

BIOINFORMATICS ANALYSIS OF THE NGS DATA

NGS workflows across targeted panels, WES, and WGS share core steps—DNA extraction, library preparation, sequencing, and analysis—but differ mainly in enrichment strategies and genomic coverage, balancing depth and coverage for specific clinical needs. Library preparation involves fragmentation, adapter ligation, and (where applicable) target enrichment, followed by high-throughput sequencing on platforms such as Illumina, generating data whose interpretation is central to downstream bioinformatics analysis.^[4]

From a clinical perspective, appropriate sequencing depth and coverage are critical to ensure reliable variant detection. WES typically requires a mean depth of ~100×, with ≥95 —98% of target regions covered at ≥20×, to compensate for capture-related variability. In contrast, WGS provides more uniform genomic coverage and can achieve reliable variant detection at a lower mean depth of ~30×, with ≥95% of the genome covered at ≥10×.

The raw data generated from sequencing must then undergo structured bioinformatics processing to enable accurate variant identification and clinical interpretation. NGS bioinformatics involves three key stages:^[5]

1. Primary: conversion of raw sequencing signals into FASTA with quality (FASTQ) reads with quality control

2. Secondary: alignment to the reference genome and variant calling to generate variant call format (VCF) files

3. Tertiary: variant annotation, filtering, and prioritization.

While the initial steps ensure data accuracy and variant detection, the most clinically meaningful stage is tertiary analysis, which focuses on variant annotation, filtering, and prioritization—integrating genomic data with phenotype and inheritance patterns to identify clinically relevant variants.

This review will primarily concentrate on the clinical variant prioritization and interpretation. Readers can refer to Flow Chart 1 and Table 1 for an overview of bioinformatics tools used across primary, secondary, and tertiary NGS data analysis.

Close

Flow Chart 1: Overview of next-generation sequencing bioinformatic analysis. Primary analysis converts raw sequencing signals into sequence reads (FASTQ files) with quality scores, followed by basic quality control and trimming. Secondary analysis involves alignment of reads to a reference genome, post-processing, and variant calling to generate variant lists (VCF files), with manual review using tools such as Integrative Genomics Viewer to confirm variant quality. Tertiary analysis includes variant annotation, integration with databases, filtering and prioritization, and final clinical interpretation based on phenotype and established guidelines. FASTQ: FASTA with quality, QC: Quality control, SNV: Single nucleotide variant, CNV: Copy number variant, VCF: Variant call format, ACMG: American College of Medical Genetics and Genomics.

Export to PPT

Bioinformatic tools used in clinical variant prioritization and interpretation can be broadly categorized into several groups based on their primary function. Numerous tools, databases, and genome browsers are available within each of these categories; however, this review will focus on selected resources that are freely accessible, widely used, and most relevant for clinical practice. The broad categories include:

1. Population variant frequency databases

2. Diseases and gene disease association databases

3. Computational prediction tools

4. Variant interpretation and classification platforms

5. Phenotype-driven and integrative tools.

Together, these tools help clinicians move from raw sequencing data to clinically meaningful interpretation by integrating genomic findings with phenotypic and functional evidence. A summary of all these resources and tools is provided in Table 2.

POPULATION FREQUENCY DATABASES

Population frequency databases are among the most important resources used in the initial filtering strategy during genetic variant assessment. Most population databases derive their data from large cohorts of apparently healthy or unaffected individuals. These databases contain information on the frequency of genetic variants across different populations and help determine whether a variant is rare or common. This distinction is crucial because rare variants are more likely to be associated with Mendelian disorders, whereas common variants present in the general population are usually benign. Thus, these resources help clinicians distinguish potentially disease-causing variants from the millions of largely benign variants present in every human genome. However, it is important to exercise caution when using population frequency data. Although pathogenic variants associated with Mendelian disorders are typically rare, most rare variants are not pathogenic. Therefore, rarity alone is insufficient to establish pathogenicity and must be interpreted alongside other lines of evidence.

The most frequently used databases are discussed below.

DISEASES AND GENE DISEASE ASSOCIATION DATABASES

DECIPHER

This database (https://www.deciphergenomics.org) contains information on anonymized, phenotype-linked genomic data from patients with rare diseases. It integrates genomic variants with detailed clinical features encoded using Human Phenotype Ontology (HPO) terms, facilitating genotype — phenotype correlation and variant interpretation. The platform contains information on a wide range of variant types (sequence variants, short tandem repeats, copy-number variants, and large structural variants), including germline and mosaic variation in the nuclear and mitochondrial genomes.^[12]

COMPUTATIONAL PREDICTION TOOLS

In silico prediction tools are computational algorithms used to estimate the potential pathogenicity of genetic variants. Most prediction algorithms integrate multiple sources of information, including evolutionary conservation, protein structure, biophysical properties, and genomic context, to assess the likelihood that a variant may disrupt gene or protein function. The two main categories of such tools are those that predict whether a missense change damages the protein’s function or structure, and those that predict whether it affects splicing.

At present, the most widely used prediction tools are ensemble or meta-predictors, based on machine-learning approaches. These tools combine outputs from multiple individual prediction algorithms along with additional genomic features—such as conservation scores and allele frequency—to generate a single aggregate pathogenicity score. Examples of commonly used tools include REVEL, a meta-predictor frequently used to assess missense variants, which assigns higher scores (typically >0.5 or >0.75) to variants with a greater likelihood of pathogenicity. Other ensemble predictors include BayesDel. More recently, deep learning-based tools such as AlphaMissense have been developed to evaluate missense variants by integrating protein structural context with evolutionary information. For variants that may affect RNA splicing, specialized tools such as SpliceAI are widely used and are considered among the most effective methods for predicting splice-altering variants.^[14] Combined annotation-dependent depletion (CADD) is another widely used in silico tool that combines multiple biological evidences to predict the deleteriousness of SNVs and small insertion/deletions (indels). Higher CADD scores (e.g., Phred-scaled score 20) indicate a higher probability that a variant is deleterious or pathogenic.^[15]

Despite their sophistication, these tools should be interpreted with caution. In silico predictions provide supporting evidence but should not be used in isolation to determine variant pathogenicity or as stand-alone diagnostic criteria.

FUNCTIONAL EVIDENCE IN VARIANT INTERPRETATION

Functional evidence provides direct insight into the biological impact of genetic variants and represents a higher level of evidence compared to in silico predictions. Experimental studies—including enzyme activity assays, electrophysiological analyses, and cell-based functional assays—can demonstrate the effect of a variant on protein function. However, unlike population and disease databases, functional data are not consolidated within a single comprehensive resource. While selected databases, such as ClinVar, may include summaries of functional studies submitted by laboratories, and specialized resources such as MaveDB (https://www.mavedb.org) provide high-throughput functional data for specific genes, the availability of such data remains limited and variable across variants. Consequently, much of the functional evidence used in clinical interpretation is derived from published studies, making literature databases such as PubMed essential for identifying experimentally validated variant effects.

VARIANT INTERPRETATION AND CLASSIFICATION PLATFORMS

After gathering supporting evidence from population databases, disease databases, in silico prediction tools, and functional evidence, the next critical step is the systematic interpretation and classification of genetic variants. To ensure consistency and clinical reliability, standardized frameworks have been developed to guide this process. The most widely adopted guidelines are those proposed by the American College of Medical Genetics and Genomics (ACMG) and the Association for Molecular Pathology (AMP), which provide a structured approach for the interpretation of sequence variants.^[16] These guidelines integrate multiple lines of evidence—including population frequency, computational predictions, functional studies, segregation data, and phenotype correlation—to classify variants into five categories: Pathogenic, likely pathogenic, VUS, likely benign, and benign. A similar standard has also been developed by ACMG and Clinical Genome Resource (ClinGen) for the interpretation and reporting of constitutional copy number variants.^[17]

To further refine variant interpretation and strengthen gene-disease validity assessment, ClinGen was established as an international initiative to systematically curate clinically relevant genes and variants (https://www.clinicalgenome.org/start). ClinGen expert panels evaluate the strength of evidence supporting gene –disease relationships and provide gene and disease-specific recommendations for variant interpretation. These curated resources help standardize variant classification across laboratories, reduce interpretative discrepancies, and improve diagnostic accuracy.^[18] As a result, the use of ClinGen-curated data contributes to higher rates of definitive molecular diagnoses and a reduction in the number of variants classified as VUS.

These guidelines provide a standardized framework for variant classification, and their effective application in clinical practice requires integration with a detailed patient phenotype.

The following section outlines bioinformatic tools that facilitate this genotype —phenotype correlation and support clinically meaningful interpretation.

PHENOTYPE-DRIVEN AND INTEGRATIVE TOOLS

A range of tools has been developed to standardize phenotypic descriptions, enable phenotype-driven diagnosis, and integrate clinical features with genomic data for effective variant prioritization. These approaches not only improve diagnostic yield but also enhance the clinical relevance of NGS findings.

APPROACH TO THE CASE SCENARIO

The initial clinical and biochemical evaluation raised suspicion for a syndromic form of diabetes. To standardize phenotypic characterization and facilitate computational analysis, key clinical features were mapped to HPO terms, including hyperglycemia, diabetic ketoacidosis, infantile onset, developmental delay, hypotonia, and hypsarrhythmia. Phenotype-driven diagnostic tools (such as Phenomizer/Face2Gene/GDDP) prioritized differential diagnoses, with developmental DEND syndrome emerging as the most likely candidate, followed by transient neonatal diabetes mellitus and other syndromic insulin resistance disorders.

Subsequently, genetic testing using WES identified a heterozygous missense variant in the KCNJ11 gene (c.602G>A; p. Arg201His). This gene encodes the Kir6.2 subunit of the ATP-sensitive potassium (K⁺ATP) channel, which plays a critical role in pancreatic β-cell insulin secretion and neuronal excitability.

A structured variant interpretation was performed in accordance with ACMG —AMP guidelines. The variant was absent in population databases such as gnomAD and IndiGenomes (PM2), supporting rarity. It had been previously reported in disease databases, including ClinVar and LOVD (PP5), suggesting prior association with disease. Multiple in silico prediction tools (REVEL, CADD, and AlphaMissense) indicated a deleterious effect (PP3). In addition, the variant is located in a functionally critical domain of the protein (PM1), and other pathogenic missense variants affecting the same codon have been reported (PM5). The gene is known to have a missense mutation mechanism (PP2). Based on this cumulative evidence (PM1, PM2, PM5, PP2, PP3, and PP5), the variant was classified as likely pathogenic.

Correlation of the genetic findings with the clinical phenotype confirmed the diagnosis of DEND syndrome due to a KCNJ11 mutation. This diagnosis had immediate therapeutic implications. The patient was transitioned from insulin therapy to oral sulfonylurea (glibenclamide), which directly targets the K⁺ATP channel defect, resulting in improved glycemic control and allowing discontinuation of insulin.

Genetic counseling was provided to the family, explaining the autosomal dominant inheritance pattern, with most cases being de novo, and a recurrence risk of <1% due to the possibility of germline mosaicism [Flow Chart 2].

Close

Flow Chart 2: Approach to the case scenario. DKA: Diabetic ketoacidosis, LOVD: Leiden Open (source) Variation Database, CADD: Combined annotation-dependent depletion, AD: Autosomal dominant, ROR: Risk of recurrence, WES: Whole exome sequencing, HPO: Human Phenotype Ontology, EEG: Electroencephalogram, MRI: Magnetic Resonance Imaging.

Export to PPT

CONCLUSION

Bioinformatic tools have become integral to the interpretation of genomic data in pediatric endocrine disorders, supporting clinicians in interpreting complex sequencing data for clinical application. Effective use of bioinformatic resources—including databases, computational tools, and integrative platforms—requires a practical understanding of their application and limitations. Familiarity with these tools empowers clinicians to critically evaluate genetic reports, correlate findings with clinical phenotypes, and make informed decisions, ultimately improving diagnostic accuracy and patient outcomes.

GLOSSARY

1. Gene panels: A technique that involves analyzing multiple genes simultaneously to identify genetic variants associated with specific diseases or conditions

2. WES: A technique that targets and sequences only the protein-coding regions (1 —2% of the genome) of an individual’s DNA, known as the exome

3. WGS: A technique that determines the entire DNA sequence, including protein-coding regions and non-coding regions of the genome

4. Variant: Alteration in the DNA sequence compared to a reference, replacing the term “mutation”

5. FASTQ: Standard text-based format for storing raw NGS data, containing both nucleotide sequences and their corresponding base quality scores

6. VCF: Standard text file format to store information about DNA sequence variations found in a sample compared to a reference genome

7. Variant annotation: Process of attaching biological, functional, and clinical information to raw genetic variants identified in a VCF file

8. Sequencing depth: Number of times a particular nucleotide is read during the sequencing process

9. Sequencing coverage: Pertains to the proportion of the genome or targeted region that has been sequenced at least once

10. MAF: The frequency at which the second most common allele (the “minor” allele) occurs in a given population for a specific single-nucleotide polymorphism

11. Sequence variant: A permanent change in the nucleotide sequence in the DNA sequence compared to a reference sequence. These changes include single-nucleotide substitutions, small deletions, insertions, duplications, and splicing changes

12. Copy number variant: A structural genomic alteration where a segment of DNA, typically larger than 1000 base pairs, is present in a variable number of copies compared to a reference genome. These include large deletions or duplications.