Singleton exomes leave the hardest question unanswered
A standard singleton whole-exome analysis can call tens of thousands of variants, and even aggressive population-frequency and ACMG-based filtering typically leaves several hundred candidates per case. Without a second and third genome to compare against — the parents' — a geneticist has no automated way to tell which of those candidates arose spontaneously in the proband and which were simply inherited from an unaffected parent. That distinction is often the difference between a shortlist a clinician can review in an afternoon and one they can't.
About the Study
A clinical genetics service running exome sequencing for pediatric patients with suspected monogenic disease piloted a trio-based version of their WES workflow across 25 proband–parent trios. The goal was to move from "here are 400 variants that pass our filters" to "here are the two or three variants most likely to explain this child's phenotype" — without adding manual review time per case.
"Singleton exomes were giving us a haystack. What we actually needed was a pipeline that knew which straws came from the parents and which ones didn't — and then ranked what was left against the phenotype we gave it."
Lead Clinical Geneticist
Client name withheld per confidentiality agreement
The Challenge
Volume and inheritance context, not sequencing quality, were the bottleneck
The client's sequencing and initial variant calling were already solid — the problem sat downstream, in interpretation:
- Standard population-frequency and quality filters alone still returned too many candidates per proband to review manually within a reasonable turnaround time.
- Without parental genotypes, the pipeline couldn't distinguish a de novo variant from a rare inherited one, or confirm compound-heterozygous inheritance in recessive disease genes.
- Phenotype context — the HPO terms describing each proband's clinical presentation — wasn't being used systematically to rank candidates, so equally "rare" variants in relevant and irrelevant genes carried the same weight.
The brief was to re-run variant calling jointly across each trio, apply inheritance-aware filtering on top of standard quality and population filters, and rank whatever remained against each proband's HPO terms — producing a shortlist rather than a spreadsheet.
The GenomeBeans Approach
GenomeBeans' WES Trio Germline pipeline processes proband and parent samples together from alignment through to a phenotype-ranked shortlist.
For this cohort, the workflow ran as follows:
-
Quality control and trimming
Raw FASTQ files from all three trio members were assessed for base quality, adapter content, duplication rate, and GC distribution, with per-sample metrics aggregated into a single report so any one weak sample in a trio was visible immediately, rather than surfacing only after alignment.read QC + adapter/quality trimming, aggregated per trio
-
Alignment and duplicate marking
Trimmed reads from proband and both parents were independently aligned to GRCh38/hg38, with PCR duplicates marked per sample ahead of joint calling. Keeping alignment per-sample (rather than merging trio reads) preserves each individual's genotype signal for the joint-calling step.per-sample alignment, GRCh38/hg38
-
Joint variant calling across the trio
Rather than calling each sample separately and comparing VCFs afterward, the pipeline calls proband and parent samples jointly in gVCF mode, then performs joint genotyping across all three — producing a single multi-sample VCF where each variant carries a genotype for every trio member.gVCF joint genotyping, trio-aware
-
Variant annotation
Every variant in the joint-called VCF was annotated for functional consequence, HGVS nomenclature, and in-silico pathogenicity scores, giving the downstream filtering steps the evidence needed to weigh each candidate rather than just count it.consequence + HGVS + pathogenicity scoring
-
Trio inheritance filtering
With genotypes available for all three family members, the pipeline classified each surviving variant by inheritance pattern — de novo (absent in both parents), autosomal recessive (homozygous in the proband, heterozygous in each parent), compound heterozygous (two different heterozygous variants in the same gene, one from each parent), or X-linked — and discarded variants that were simply inherited from an unaffected parent with no plausible disease relevance.de novo / recessive / compound-het / X-linked logic
-
HPO-driven prioritization
Surviving candidates were ranked against each proband's Human Phenotype Ontology terms, scoring variant-gene-phenotype fit so that a rare, well-supported variant in a gene with no connection to the reported phenotype ranked below a more moderately-supported variant in a gene tied directly to the clinical presentation.HPO-based variant-gene-phenotype scoring
Conclusion
The pilot confirmed what the client suspected: the bottleneck was never sequencing quality, it was the absence of inheritance and phenotype context in the filtering step. By joint-calling each trio and layering de novo/recessive/compound-het logic on top of standard quality and population filters, the pipeline turned an average of roughly 11,000 joint-called variants per trio into a shortlist of about four candidates — small enough to review with the proband's actual clinical picture in hand, not just a spreadsheet of allele frequencies.
For the client, that means a workflow that scales the same way regardless of case volume: sequencing quality stays the responsibility of the wet lab, and the pipeline's job — narrowing thousands of variants down to a phenotype-ranked few — happens automatically, the same way, for every trio that comes through.