Whole Exome Sequencing · Research & Diagnostics

Decode the coding genome. Discover disease variants.

End-to-end Whole Exome Sequencing and analysis for human and eukaryotic genomes. Cost-effective germline variant discovery targeting the protein-coding regions where most known disease-causing variants reside.

Request a Quote → See the Pipeline
1–2%
of the genome, majority of disease variants
>99%
Mean primary alignment rate
~21K
Coding-region variants per sample
2
Independent callers: GATK + DeepVariant
About the Service

Precision variant discovery made accessible

Nucleome Informatics offers end-to-end Whole Exome Sequencing (WES) and data analysis services for human and other eukaryotic genomes — a cost-effective alternative to whole genome sequencing for germline variant discovery in research, translational, and diagnostic settings.

Our exome pipelines are built on current best-practice tools and are run and benchmarked using two independent variant-calling strategies — GATK HaplotypeCaller and DeepVariant — giving clients the flexibility to choose the pipeline that best fits their study design, or to benchmark both against each other.

  • Raw sequence quality control, adapter trimming, and quality filtering
  • Alignment to the reference genome (BWA-MEM, hg38/GRCh38 or client-specified)
  • Duplicate marking and base quality score recalibration (GATK best practices)
  • Germline SNP/indel calling with GATK HaplotypeCaller and/or DeepVariant
  • Hard-filtering of variants against established quality thresholds
  • Functional and clinical annotation with ANNOVAR (ClinVar, gnomAD, dbNSFP)
  • Sample-wise and cohort-level summary reporting

Dual-Caller Advantage

Clients can directly benchmark caller performance on their own data rather than relying on published comparisons. Both GATK HaplotypeCaller and DeepVariant run on the same aligned, recalibrated BAM files — delivering reproducible, high-quality variant sets you can trust.


Clinically Relevant Annotation

Filtered variants are annotated against ensGene, avsnp151, dbNSFP42c, ClinVar, gnomAD (exome), and population-frequency databases — ready for research or clinical interpretation.

Learning Path

From first principles to clinical-grade interpretation

New to exomes? Start with Foundations. Running projects or reviewing results? Jump to Practitioner or Expert. Each level builds on the one before it, so you can read this page as a short course or dip in where you need to.

The big idea in plain language: a whole exome reads the parts of your DNA that are used to build proteins, and looks for spelling differences that may explain disease.

LEVEL 1

Genes are recipes, exons are the words

A gene is a recipe for a protein. Only some stretches of it, the exons, are actually translated. The rest (introns) is cut out. All exons together make up the exome.

LEVEL 1

Small region, big payoff

The exome is only about 1–2% of the genome, yet it is where most known disease-causing variants are found. That is why sequencing it is a cost-effective first look.

LEVEL 1

Probes fish out the exons

DNA is cut into fragments, then short probes that match exons are used like hooks to pull those fragments out of the mix. Only the pulled-out fragments are sequenced.

LEVEL 1

Reads are puzzle pieces

The sequencer produces millions of short reads. Software maps each one to its place on the human reference genome (hg38) so the pieces line up into a picture of your exome.

LEVEL 1

A variant is a difference

Where your sequence differs from the reference, that is a variant. It can be a single letter change (SNV) or a small insertion or deletion (indel), and you can carry one copy (heterozygous) or two (homozygous).

LEVEL 1

Annotation adds meaning

A list of coordinates means little. Annotation adds which gene is affected, what changes in the protein, how common the variant is in the population, and what clinical databases say about it.

For scientists planning a project or reading a report: how depth, calling and filtering translate into a short, trustworthy list of variants.

LEVEL 2

Depth is not the whole story

Mean depth tells you the average. What matters is how many target bases reach a usable depth (often judged at 20×), which depends on capture uniformity. Use the coverage simulator below to see why.

LEVEL 2

Two ways to call variants

GATK HaplotypeCaller re-assembles reads locally around candidate sites. DeepVariant turns read pileups into images and classifies them with a neural network. Running both on the same BAM lets you cross-check difficult sites.

LEVEL 2

Hard filters remove weak calls

Metrics such as QD (quality by depth), FS (strand bias), MQ (mapping quality) and SOR (strand odds ratio) flag unreliable calls. Commonly used GATK starting points are QD < 2 and FS > 60 for SNVs, tuned per project.

LEVEL 2

Rare variants matter most

Variants common in healthy populations rarely cause rare disease. Filtering against gnomAD and similar resources removes most of the noise, leaving a few hundred rare, protein-altering candidates.

LEVEL 2

Family structure is a filter

Sequencing a child with both parents (a trio) shows which variants arose de novo or segregate with disease. A published meta-analysis found trios gave roughly twice the odds of a diagnosis compared with testing a child alone.

LEVEL 2

Sanity checks catch problems early

Cohort-level checks such as the transition/transversion (Ti/Tv) ratio, sex and relatedness checks, and contamination estimates show whether a sample can be trusted before anyone interprets it.

For clinical geneticists and computational biologists: classification frameworks, computational evidence, structural limits of the assay, and where the field is moving.

LEVEL 3

ACMG/AMP classification

The 2015 ACMG/AMP guidelines combine 28 evidence criteria into five classes from pathogenic to benign. Points-based reformulations make the evidence weighting explicit and reproducible.

LEVEL 3

Computational evidence

Predictors such as REVEL, CADD, SpliceAI and AlphaMissense contribute supporting evidence. AlphaMissense scored about 71 million possible missense variants, calling roughly 32% likely pathogenic and 57% likely benign, but predictions are not diagnoses on their own.

LEVEL 3

What an exome cannot see

Exome capture is designed for coding exons. Deep intronic variants, most structural variants, repeat expansions and uneven-coverage regions are poorly or not detected, and copy-number calling from exome depth has limited resolution.

LEVEL 3

Population resources at scale

gnomAD v4.0 aggregates 807,162 individuals (730,947 exomes and 76,215 genomes), enabling much finer frequency filtering and constraint-based interpretation, ideally matched to ancestry.

LEVEL 3

Reanalysis pays off

Knowledge about genes and variants keeps growing. A literature review of 27 studies reported a median diagnostic yield of about 15% from reanalysis, with a median gap of about 22 months after the original test.

LEVEL 3

Secondary findings and consent

Clinical exome programmes may report medically actionable variants in a curated gene list (ACMG SF v3.3, 2025). Whether and how to report them is a consent and policy decision made with the clinical team.

Exome
All protein-coding exons of the genome, about 1–2% of total DNA.
Read depth
How many reads cover a base. Higher depth gives more confident genotypes.
Uniformity
How evenly depth is spread across targets; poor uniformity leaves gaps.
VCF
Standard text file listing variants, genotypes and quality metrics.
Heterozygous
One reference copy and one variant copy of a gene.
De novo
A variant present in the child but absent from both parents.
VUS
Variant of uncertain significance: evidence is insufficient to call it harmful or harmless.
Allele frequency
How often a variant occurs in a reference population such as gnomAD.
The Science

How an exome is captured and read

Whole exome sequencing enriches the protein-coding portion of the genome before sequencing, concentrating reads where disease-causing variants are most often found.

Figure 1 From genome to genotype schematic
1 Genome2 Library3 Hybridise4 Capture5 Sequence exon intron (non-coding) fragments + adapters B biotinylated probe no exon, no probe strept-avidin washedaway variant (3 of 6 reads → het) Exons make up ~1–2%of the genome DNA sheared; adaptersadded to both ends Probes pair withexon-containing fragments Beads pull probes out;the rest is washed away Reads align to hg38;variants are called
Figure 1 | Hybridisation-capture exome sequencing. (1) Exons, the protein-coding segments, form a small fraction of the genome. (2) DNA is fragmented and adapters are ligated for sequencing. (3) Biotinylated probes complementary to exons hybridise to matching fragments. (4) Streptavidin beads pull down probe-bound fragments while off-target DNA is washed away. (5) Enriched fragments are sequenced and aligned; positions where a fraction of reads differ from the reference reveal variants, and the ratio of alternate to reference reads indicates the genotype.
Figure 2 How small is the exome? service stat
Human genome ≈ 3.1 billion basescoding exons ≈ 1–2%
~20,000protein-coding genes
1–2%of the genome sequenced
~21Kcoding variants per sample in our case study
Figure 2 | Scale of the target. Exome capture concentrates sequencing on a sliver of the genome, which is why a high depth is affordable. Green ticks illustrate exon positions and are not to scale.
Figure 3 What kinds of variants are in a typical exome? illustrative
Synonymous 49% Missense 44%
Synonymous (no amino-acid change) Missense (one amino acid changed) In-frame indels, splice region, other (~5%) Predicted loss-of-function (~1–2%)

Most variants in any exome are harmless. The ones that matter for disease are usually the small, rare, protein-altering subset on the right, which is why filtering and annotation are so important.

Figure 3 | Composition of coding variants. Approximate proportions for a typical human exome, shown to convey scale; exact fractions vary by ancestry, capture kit and filters.
Sample Requirements

What we need to get started

Standard input requirements for a WES project. Extracted FASTQ files are also accepted if you prefer to handle library preparation and sequencing externally.

Parameter Requirement
Sample type Genomic DNA (blood, saliva, or tissue-extracted); extracted FASTQ also accepted
Minimum quantity As per project design — contact us for specific thresholds
Minimum concentration As per project design — contact us for specific thresholds
Purity (A260/280) Standard high-quality DNA purity recommended
Integrity High molecular weight DNA preferred
Recommended sequencing depth Project-dependent; typically high-coverage for clinical-grade calls
Read configuration Paired-end (standard Illumina chemistry)
Shipping / storage Ship on ice / dry ice as appropriate; store at −20 °C or lower
Sample sheet Required — sample IDs, metadata, and any special instructions

Exact quantity, concentration, and depth requirements vary by project scope. Contact us at info@nucleomeinfo.com for a tailored sample checklist.

Quantify by fluorometryDye-based quantification reflects double-stranded DNA better than absorbance alone, which also reads RNA and contaminants.
Check purity ratiosClean DNA typically shows A260/280 near 1.8 and A260/230 between about 2.0 and 2.2.
Protect integrityAvoid repeated freeze–thaw cycles and harsh extraction; high-molecular-weight DNA gives better, more even libraries.
Send rich metadataFamily structure, affected status and phenotype terms in the sample sheet drive filtering and interpretation.
Workflow

From raw reads to annotated variants

Every sample moves through the same quality-controlled pipeline, whether variants are called with GATK, DeepVariant, or both.

Figure 4 QC-gated pipeline with dual-caller branch infographic
STAGE 01

Raw data QC

FastQC reports base quality and adapter content; Fastp removes adapters and trims low-quality bases.

FASTQHTML / JSON
Q30 · adapters
STAGE 02

Alignment

Clean paired-end reads mapped to hg38 (or client build) with BWA-MEM.

BAM
primary alignment rate
STAGE 03

Duplicates & BQSR

MarkDuplicates flags PCR and optical duplicates; BaseRecalibrator and ApplyBQSR recalibrate quality scores.

BAM + index
duplicate rate
STAGE 04

Variant calling

Same recalibrated BAM, two independent strategies.

GATK HaplotypeCallerDeepVariant v1.6.0
VCFgVCF
Ti/Tv · call counts
STAGE 05

Filtration

Hard filters (QD, FS, MQ, SOR and related metrics) remove low-confidence calls.

filtered VCF
PASS / FAIL flags
STAGE 06

Annotation

ANNOVAR adds ensGene, avsnp151, dbNSFP42c, ClinVar, gnomAD and population databases.

annotated VCFTXT
annotation coverage
STAGE 07

Reporting

Sample-wise and cohort summary statistics compiled for delivery.

summary report
cohort sanity checks
STAGE 08

Secure delivery

All files delivered through secure cloud storage with role-based access.

shared drive
access control
Figure 4 | Decision points. Amber gates mark where metrics are reviewed before a sample proceeds. Running both callers on the same aligned, recalibrated BAM allows direct benchmarking on your own data.
1

Raw Data QC

FastQC assesses base quality & adapter content; Fastp removes adapters and trims low-quality bases.

2

Alignment

Clean reads mapped to hg38 (or client build) using BWA-MEM, a Burrows-Wheeler Transform aligner.

3

BQSR & Duplicates

GATK MarkDuplicates flags PCR/optical duplicates; BaseRecalibrator + ApplyBQSR recalibrate scores.

4

Variant Calling

SNPs & small indels called with GATK HaplotypeCaller and/or DeepVariant (v1.6.0).

05

Variant filtration — Hard filters (QD, FS, MQ, SOR, and related metrics) remove low-confidence calls.

06

Annotation — ANNOVAR against ensGene, avsnp151, dbNSFP42c, ClinVar, gnomAD, and population databases.

07

Reporting — Sample-wise and cohort summary statistics compiled for delivery.

08

Secure delivery — All files delivered via secure cloud storage with role-based access.

Quality Control in Depth

Reading the diagnostic plots like a reviewer

Good exome studies show their quality control. These are the checks that tell you whether reads, coverage and variant calls can be trusted before anyone interprets a single variant.

Figure 5 Per-base sequence quality simulated
Figure 5 | Trimming rescues read ends. Base quality (Phred score) typically falls toward the end of a read. Adapter and quality trimming lifts the mean and narrows the spread. In our 5-sample case study the mean Q30 rose from 92.7% (raw) to 96.6% (clean).
Figure 6 Cohort sanity check: Ti/Tv ratio simulated
Figure 6 | Transition/transversion ratio. Real variants are enriched for transitions, giving an expected ratio near 2.8 in exomes (about 2.1 genome-wide). Random errors push the ratio toward 0.5, so samples that fall well below the cohort are flagged for contamination or poor library quality.
Figure 7 Interactive coverage simulator model
Mean target depth
Figure 7 | Why mean depth is not enough. The curves show the fraction of target bases covered at least x-fold. A perfectly even library would follow the Poisson curve (green). Real capture is uneven (over-dispersed, orange), so a meaningful share of bases falls below 20× even at high mean depth. Drag the slider to explore. Clinical exome programmes commonly aim for a mean depth around 100× with most target bases at 20× or more; your project's depth is agreed with our team.
Figure 8 Two callers, one BAM simulated
Figure 8 | Caller concordance. SNVs are largely shared between GATK HaplotypeCaller and DeepVariant, while indels show proportionally more disagreement. Benchmarking both on your own data, as our dual-caller design allows, shows which calls are robust and which need review.
Figure 9 How DeepVariant sees a variant schematic
Read pileup as an image bases, quality, strand,mapping quality as channels Deep neural network convolutionlayersclassify Genotype probability hom-refhethom-alt Learns variant patterns from labelled truth data rather than relying on hand-built statistical rules
Figure 9 | Variant calling as image classification. DeepVariant encodes the reads around a candidate site as a multi-channel image and a convolutional neural network outputs genotype probabilities. Versions are trained for specific data types, including Illumina whole-exome sequencing; we run v1.6.0.
From Variants to Answers

Finding the needle: filtering and interpretation

A single exome contains tens of thousands of variants. Interpretation is a sequence of filters and evidence weighting that narrows them to a handful of candidates. Nucleome delivers annotated, quality-filtered variants (ClinVar, gnomAD, dbNSFP and more) as the input to this process.

Figure 10 The filtering funnel illustrative
Coding-region variants called≈21,150 per sample in our case study
~21,150
Protein-alteringmissense, loss-of-function, splice, indels
~10,800
Rare in populationallele frequency below ~1% in gnomAD
~1,100
Predicted damagingconservation, REVEL, CADD, AlphaMissense, SpliceAI
~150
Fits inheritance and phenotypede novo, recessive, X-linked; gene-phenotype match
~8
Candidate diagnostic variant(s)reviewed against ACMG/AMP criteria
1–3
Figure 10 | Each filter removes noise. Bar lengths are on a logarithmic scale. Only the first count comes from our case-study data; the later steps are typical orders of magnitude, shown for explanation, and real values depend on the filters, phenotype and family structure.
Figure 11 Inheritance patterns that narrow candidates schematic
De novo (dominant)Autosomal recessiveCompound heterozygousX-linked recessive 0/00/00/1 (new) 0/10/11/1 A/refB/refA and B X/Y refXv/XXv/Y Absent from both parentsTwo copies, one from each carrierTwo different variants in one geneCarrier mother passes to son unaffectedaffectedcarriersquares = male, circles = female
Figure 11 | Family-based filtering. Genotype patterns across a trio sharply reduce the candidate list: de novo variants are absent from both parents; recessive disease needs two altered copies, either homozygous or as two different variants in the same gene (compound heterozygous); X-linked variants pass from carrier mothers to affected sons. Genotypes follow VCF notation (0 = reference, 1 = variant).
Figure 12 ACMG/AMP variant classification guideline framework
Pathogenic≥ 10 points
Likely pathogenic6–9 points
Uncertain (VUS)0–5 points
Likely benign−1 to −6 points
Benign≤ −7 points
Evidence for pathogenicity. Very strong (PVS1, e.g. null variant in a loss-of-function-intolerant gene), strong (PS), moderate (PM, e.g. absent from population databases, hotspot) and supporting (PP, e.g. computational predictions, segregation).
Evidence for benign impact. Stand-alone (BA1, allele frequency too high for the disorder), strong (BS) and supporting (BP, e.g. computational evidence of no impact, synonymous with no splice effect).
Population data

gnomAD v4.0: 807,162 individuals. Variants too common for the disease argue for benign.

Computational predictors

REVEL, CADD, SpliceAI, AlphaMissense. Supporting evidence only; predictors are correlated, not independent.

Clinical and functional data

ClinVar assertions, functional assays, segregation, de novo status and phenotype specificity.

Figure 12 | Combining evidence. The 2015 ACMG/AMP guidelines define 28 criteria that combine into five classes. The point thresholds shown follow the published points-based (Bayesian) reformulation, where supporting, moderate, strong and very strong evidence carry increasing weight. Classification is an expert process that uses phenotype and family information; Nucleome's deliverables supply the annotated evidence layers.
Annotation sourceWhat it addsHow it is used
ensGeneGene models and variant consequence (missense, nonsense, splice, synonymous)Decides which variants alter a protein
avsnp151dbSNP identifiers for known variantsLinks calls to published records
dbNSFP42cAggregated pathogenicity and conservation scores from many predictorsPrioritises likely-damaging missense variants
ClinVarClinical significance assertions submitted by laboratories and expertsFlags known pathogenic or benign variants
gnomAD (exome)Allele frequencies across large reference populationsRemoves variants too common to cause rare disease
Deliverables

Everything you need, ready to use

At project close, clients receive a complete data package — from raw reads to clinically annotated variant tables — via a secure shared drive link.

📁

Raw & Clean Sequence Data

FASTQ files (raw and adapter/quality-trimmed) for every sample.

📊

QC Reports

FastQC / Fastp HTML and JSON reports per sample for full transparency.

🧬

Alignment Files

Duplicate-marked, recalibrated BAM files with index — analysis-ready.

🔍

Variant Calls

Per-sample and/or joint-genotyped VCFs (raw and hard-filtered).

📋

Annotated Variants

ANNOVAR-annotated multi-database VCF and Excel-ready tab-delimited text files.

📈

Summary Report

Sample-wise QC, alignment, variant-calling, and annotation statistics.

Case Studies

Proven performance on real cohorts

Real production metrics from human exome cohorts processed through our standard pipeline, alongside a published South Indian cohort study showing what exome sequencing delivers at population scale.

Germline Variant Discovery

5-Sample Human Exome Cohort · DeepVariant Pipeline

Mean raw data Q30 92.7%
Mean clean data Q30 96.6%
Mean primary alignment rate >99%
Mean coding-region variants / sample ~21,150
Mean high-confidence variants / sample ~24,200

All samples cleared quality thresholds after trimming, achieved near-complete alignment to hg38, and yielded high-confidence, clinically annotated variant call sets delivered as raw and annotated VCFs alongside a full summary report.

Variant-Caller Benchmarking

DeepVariant vs. GATK HaplotypeCaller · 40+ Exomes

Primary alignment rate >99%
Post-filtering Q30 >96%
Callers supported Both (parallel)
Same BAM input Yes
Client benchmarking Enabled

Because our pipeline supports both DeepVariant and GATK HaplotypeCaller on the same aligned, recalibrated BAM files, clients can directly benchmark caller performance on their own data — delivering reproducible, high-quality variant sets with either (or both) callers.

South Indian Retinitis Pigmentosa Cohort published data

109 Probands · Clinical Exome Workflow · Aravind Eye Hospital, Ophthalmology Science 2026

Molecular diagnostic yield 45.9% (50/109 probands)
Rare variants identified 186 across 92 retinal disease genes
Pathogenic / likely pathogenic 26 / 33 variants
Inheritance of solved cases 88% autosomal recessive
Most implicated genes EYS · USH2A · ADGRV1

Published evidence for what a clinical exome workflow delivers in an Indian population. In this consanguinity-enriched cohort, autosomal recessive disease dominated and consanguineous cases were exclusively homozygous (17/17) — yet 68% of non-consanguineous recessive cases were homozygous too. Recurrent alleles included the established South Asian founder variant MFSD8 c.1361T>C and candidate founder alleles in EYS (c.4321C>T) and ADGRV1 (c.14329C>T). Region-specific databases such as IndiGen are essential for interpreting frequencies in cohorts like this. Study by Jaju et al., not a Nucleome project; see the reference below.

The Frontier

What exome sequencing can do, and where it is heading

Published evidence shows that exome sequencing is a first-line diagnostic test for many suspected genetic conditions. Here is how it compares with other approaches, and how new AI tools and larger reference datasets are improving interpretation.

Figure 13 Diagnostic yield in children with suspected genetic disease published data
Genome sequencing
41%
Exome sequencing
36%
Chromosomal microarray
10%
Trios roughly doubled the odds of a diagnosis compared with sequencing the child alone, and the difference between genome and exome yield was not statistically significant.
Figure 13 | Exome vs genome vs microarray. Pooled diagnostic yields from a meta-analysis of 37 studies covering 20,068 children (Clark et al., npj Genomic Medicine, 2018). Newer cohorts and methods may differ.
Figure 14 AI-scored missense variants published data
32% 11% 57%
Likely pathogenic Uncertain Likely benign
71 million possible missense variants were scored by AlphaMissense (Science, 2023). Such scores help prioritise candidates, but are used as supporting rather than stand-alone evidence under ACMG/AMP.
Figure 14 | Proteome-wide missense prediction. Distribution of AlphaMissense classes across all possible human missense variants as reported by its developers. Many variants remain uncertain, which is why functional and clinical evidence still matter.
Table 1 Choosing the right test
ApproachWhat it coversStrengthsLimits
Gene panelA curated list of genes for one conditionHighest depth on few genes, simplest interpretationMisses genes outside the list; needs re-design as knowledge grows
Whole exome (WES)All protein-coding exons (~1–2% of the genome)Hypothesis-free, cost-effective, re-analysable as knowledge growsLimited for deep-intronic, structural and repeat variants; coverage gaps
Whole genome (WGS)Coding and non-coding genomeMore even coverage; detects structural variants and non-coding changesHigher cost and data volume; more variants to interpret
Chromosomal microarrayLarge copy-number changesEstablished for large deletions and duplicationsCannot see single-nucleotide or small indel variants
Table 1 | Test selection. The best choice depends on the clinical question, sample availability and budget. Exome sequencing is often a strong first, hypothesis-free step for suspected genetic disease.
TRIOS

Family-based sequencing

Sequencing parents alongside the child separates inherited from de novo variants and substantially raises diagnostic yield.

REANALYSIS

Re-reading old data

As new gene-disease links appear, re-analysing stored exome data can resolve previously unsolved cases. A review of 27 studies reported a median yield of about 15%.

COPY NUMBER

CNVs from exome depth

Read-depth methods can infer larger deletions and duplications from exome data, within the limits of capture design, and follow ACMG/ClinGen CNV interpretation standards.

SPLICING

Predicting splice impact

Deep-learning models such as SpliceAI score whether variants disrupt splicing, and RNA sequencing can confirm suspected splice variants in the right tissue.

POPULATIONS

Ancestry-aware frequencies

Large resources like gnomAD v4.0 improve rare-variant filtering, but allele frequencies should be matched to the ancestry of the individual being tested.

DEEP LEARNING

Better variant callers

DeepVariant has evolved from an Illumina-genome tool to models for exomes, long reads and trios, improving accuracy on difficult sites.

Scope note. Nucleome's WES service covers the workflow and deliverables described above, ending at quality-filtered, annotated variants and summary reports. The interpretation framework and frontier methods on this page are shown to place the service in its scientific context; contact our team to discuss extended or project-specific analyses.
Why Nucleome

Service highlights

Built for research, translational, and diagnostic exome projects — with transparency and reproducibility at every stage.

Dual Variant-Calling

GATK and DeepVariant, benchmarked side by side on the same data — choose one or run both.

Best-Practice Pipelines

Built on FastQC, Fastp, BWA-MEM, GATK, DeepVariant, and ANNOVAR — fully reproducible.

Clinical Annotation

ClinVar, gnomAD, dbNSFP, and population-frequency databases for actionable insights.

Transparent Reporting

Sample-wise QC and summary statistics at every stage — no black boxes.

Secure Cloud Delivery

Raw, intermediate, and final files delivered via secure cloud with role-based access.

Experienced Team

Bioinformatics specialists supporting research, translational, and diagnostic projects.

Questions

Frequently asked, in plain language

From the basics to the finer points of interpretation.

basicWhat is whole exome sequencing?
It reads all the protein-coding parts of your DNA, the exons, which make up about 1–2% of the genome but harbour most known disease-causing variants. It is a cost-effective alternative to whole genome sequencing.
basicDo I need to send DNA, or can I send data I already have?
Either. We accept genomic DNA from blood, saliva or tissue, and we also accept extracted FASTQ files if you handle library preparation and sequencing elsewhere.
basicWhat will I receive?
Raw and clean FASTQ, QC reports, recalibrated BAM files with index, raw and hard-filtered VCFs, ANNOVAR-annotated variant tables and a summary report, delivered through secure cloud storage.
intermediateWhy run both GATK and DeepVariant?
They use different strategies: GATK HaplotypeCaller re-assembles reads locally, while DeepVariant classifies pileup images with a neural network. Running both on the same BAM lets you benchmark them on your data and review sites where they disagree, which are most often indels.
intermediateHow much sequencing depth do I need?
It depends on the study design. What matters is not only the mean depth but the fraction of target bases covered at a usable level (commonly 20× or more). Clinical-grade calling typically uses high coverage, and our team agrees the depth with you for each project. Try the coverage simulator above to see the effect.
intermediateDoes the report tell me which variant causes the disease?
The deliverables give you quality-filtered, annotated variants (gene, consequence, population frequency, ClinVar and predictor scores). Deciding which variant explains a phenotype is an interpretation step that combines this evidence with clinical and family information, following frameworks such as ACMG/AMP.
advancedWhat can't an exome detect?
Exome capture targets coding exons, so variants deep in introns, most structural variants, repeat expansions and poorly captured regions are missed or unreliable. If an exome is negative, genome sequencing, copy-number analysis, RNA sequencing or reanalysis may be appropriate next steps.
advancedHow reliable are AI pathogenicity predictors?
They are valuable for prioritising candidates, and AlphaMissense classified about 89% of possible missense variants as likely pathogenic or likely benign. But predictors are correlated with one another and with training data, so under ACMG/AMP they provide supporting evidence rather than a verdict.
advancedWhich reference population should I filter against?
gnomAD v4.0, with 807,162 individuals, is the standard resource, but frequencies differ across ancestries, so use ancestry-matched or global filtering allele frequencies and add local population data when available.
References

Key literature & resources

Selected sources behind the methods and reference values on this page. Figures labelled "simulated", "schematic" or "illustrative" are generated for this page, not client data.

  1. Richards S, et al. Standards and guidelines for the interpretation of sequence variants: a joint consensus recommendation of the ACMG and AMP. Genetics in Medicine 2015;17:405–424.
  2. Tavtigian SV, et al. Fitting a naturally scaled point system to the ACMG/AMP variant classification guidelines. Human Mutation 2020;41:1734–1737.
  3. Clark MM, et al. Meta-analysis of the diagnostic and clinical utility of genome and exome sequencing and chromosomal microarray in children with suspected genetic diseases. npj Genomic Medicine 2018;3:16. Link
  4. Jaju S, Sivagnanam J, Menon P, et al. Whole-Exome Sequencing in a Consanguinity-Enriched South Indian Retinitis Pigmentosa Cohort: Diagnostic Yield and Molecular Spectrum. Ophthalmology Science 2026. PubMed
  5. Evaluating systematic reanalysis of clinical genomic data in rare disease from single center experience and literature review. Molecular Genetics & Genomic Medicine 2020. Link
  6. Chen S, et al. A genomic mutational constraint map using variation in 76,156 human genomes (gnomAD v4). Nature 2024;625:92–100. Release notes: gnomAD v4.0
  7. Karczewski KJ, et al. The mutational constraint spectrum quantified from variation in 141,456 humans. Nature 2020;581:434–443.
  8. Cheng J, et al. Accurate proteome-wide missense variant effect prediction with AlphaMissense. Science 2023;381:eadg7492. Summary
  9. Jaganathan K, et al. Predicting splicing from primary sequence with deep learning (SpliceAI). Cell 2019;176:535–548.
  10. Ioannidis NM, et al. REVEL: an ensemble method for predicting the pathogenicity of rare missense variants. American Journal of Human Genetics 2016;99:877–885.
  11. Rentzsch P, et al. CADD: predicting the deleteriousness of variants throughout the human genome. Nucleic Acids Research 2019;47:D886–D894.
  12. Poplin R, et al. A universal SNP and small-indel variant caller using deep neural networks. Nature Biotechnology 2018;36:983–987.
  13. Kolesnikov A, et al. DeepVariant over the years. Google Research, 2021. Link
  14. Van der Auwera GA, et al. From FastQ data to high-confidence variant calls: the Genome Analysis Toolkit best practices pipeline. Current Protocols in Bioinformatics 2013;43:11.10.
  15. Li H. Aligning sequence reads, clone sequences and assembly contigs with BWA-MEM. arXiv:1303.3997, 2013.
  16. Chen S, et al. fastp: an ultra-fast all-in-one FASTQ preprocessor. Bioinformatics 2018;34:i884–i890.
  17. Wang K, Li M, Hakonarson H. ANNOVAR: functional annotation of genetic variants from high-throughput sequencing data. Nucleic Acids Research 2010;38:e164.
  18. Zook JM, et al. An open resource for accurately benchmarking small variant and reference calls. Nature Biotechnology 2019;37:561–566.
  19. Riggs ER, et al. Technical standards for the interpretation and reporting of constitutional copy-number variants: ACMG and ClinGen. Genetics in Medicine 2020;22:245–257.
  20. ACMG SF v3.3 list for reporting of secondary findings in clinical exome and genome sequencing. Genetics in Medicine 2025. PubMed
  21. Landrum MJ, et al. ClinVar: improving access to variant interpretations and supporting evidence. Nucleic Acids Research 2018;46:D1062–D1067.