GenomeArc Launches Horizon 3.0, a Multimodal, Assembly-Based Tertiary Analysis Platform Taking Clinical Genetics Beyond the Reference

Mississauga, Ontario

Horizon^TM 3.0 is the first genome tertiary analysis platform to interpret assembled genomes, alongside long- and short-read data. By reading each person’s genome beyond a single reference, it detects more pathogenic and clinically relevant variants, helps resolve previously unsolved cases, and deciphers complex somatic mutations in tumours that reference-based methods miss. And by bringing the patient’s own single-cell and proteomic data into the same platform, it connects every variant to the biology behind it.

A four-year-old has seizures and delayed speech. Her exome is negative. So is her genome. Her family leaves without a diagnosis. Elsewhere, a tumour’s driver mutation sits in DNA the reference doesn’t contain and is never captured.

Often the answer was in the data all along. Short reads miss large repeats and insertions; long reads reach them [1]. Some variants lie in DNA the reference doesn’t contain at all [2–4]. Others are found but misclassified, because public databases underrepresent many populations [5].

Horizon^TM 3.0 follows the case all the way: short reads first, then long reads, then Phaser, which reads the patient’s assembled genome to find variants beyond the reference. At any step, single-cell and proteomic data from the patient or the tumour can add the biological evidence that DNA alone can’t give.

One platform for rare disease and cancer, from first test to last resort, with a classified result in under 15 minutes.

Long reads are here, but interpretation tools are behind

PacBio HiFi and Oxford Nanopore are moving into clinical use. Accuracy has improved, costs are falling, and the data are phased, so each variant can be assigned to the maternal or paternal copy [6].

Interpretation has not kept pace. Many tools were built for short-read data. They treat structural variants and repeats as add-ons, ignore phase, and rely on allele frequencies from short-read cohorts of mostly European ancestry. Labs pay for richer data, then lose much of it at the interpretation step.

Horizon^TM 3.0 treats long read as a primary input, while short-read workflows stay unchanged.

A multimodal platform for long and short reads

Horizon^TM 3.0 runs long-read and short-read VCFs through the same interpretation workflow, so labs don’t need a second toolchain for long-read data. Four things make this work:

• Long-read input. Horizon^TM reads phased Oxford Nanopore and PacBio HiFi VCFs, including repeats and regions that short reads miss.
• Structural variants and repeats as first-class variants. SNVs, indels, structural variants, insertions, inversions and repeat expansions are all weighed against the same ACMG rules, using public and proprietary evidence.
• Population-matched evidence. Frequencies come from long-read genomes in 10+ underrepresented populations, so common local variants no longer look falsely rare.
• Multimodal integration. Labs can upload single-cell transcriptomic and proteomic data from rare disease patients or from tumours and view it alongside reference tissue atlases in the same portal.

Rare disease: seeing the variants that hide in plain sight

For rare and inherited disease, Horizon^TM takes a long- or short-read VCF and a clinical note, ranks variants automatically under ACMG guidelines, and generates a report for expert review, with the evidence trail behind each call.

Singularity, Horizon’s rare disease module, takes phased PacBio HiFi and Oxford Nanopore files directly and classifies everything in them under ACMG guidelines:

• Structural variants: insertions, inversions, deletions and duplications, classified, not just listed.
• Repeat expansions and tandem repeats, measured rather than inferred.
• Phase: compound-heterozygous variants confirmed from the patient’s own reads.
• Homozygosity modelling for families where parents are related.
• Non-coding regions: 34,000+ lncRNA genes and regulatory elements.

When a variant of uncertain significance needs more evidence, the patient’s own single-cell or proteomic data can show whether the gene is abnormally expressed in the affected cells, supporting or questioning its likely impact. A variant in a gene that is highly active in the disease-relevant tissue rises up the review.

Each call links to its full evidence trail. Our copy-number algorithm scored highest against AnnotSV, ClassifyCNV and ISV in a peer-reviewed benchmark (AUC 0.93).

Oncology: comprehensive tumour profiling, not a list of mutations

A tumour carries many somatic variants, but an oncologist needs a short list of answers. Which changes matter, which therapies do they point to, and what does the wider biology say about where the disease is heading?

Spectra, Horizon’s oncology module, starts from tumour and matched-normal VCFs. Each somatic variant is tiered under AMP/ASCO/CAP across 27 malignant cancer types, with FDA-approved drug matches and active clinical trials attached to the call.

Spectra then builds a full tumour profile that answers the questions oncologists actually ask:

• Is immunotherapy worth considering? Tumour mutational burden is reported per megabase, normalised to the coding footprint of the assay that produced the data.
• Is DNA repair failing? A clear MSI-high or MSS call and an HRD score, with BRCA1, BRCA2 and other repair-gene variants, flag checkpoint-inhibitor and PARP-inhibitor relevance.
• What shaped this tumour? Computed mutational signatures point to the underlying processes and the targeted options that follow. Our signature calls match the benchmark at 0.9901 cosine similarity on the SBS96 profile.
• Is resistance on the way? Clonal structure, resolved down to 0.05 VAF, shows which minor clones are expanding.
• Is the driver actually active? Tumour-derived single-cell and proteomic data show whether a mutated gene is expressed, and in which cell populations, adding functional evidence to the genomic call.

Spectra support the assays labs already run, including TruSight Oncology 500, FoundationOne CDx, MSK-IMPACT, Tempus xT, Oncomine and custom panels. For hereditary cancer, germline variants are classified to ACMG/AMP alongside the tumour profile, using ClinVar, gnomAD, GenomeArc’s multi-ethnic frequencies and the patient’s family history.

Assembly-based analysis with Horizon^TM Phaser: reading each copy of the genome on its own

Most analyses start by aligning reads to a reference genome. But a standard reference misses what it doesn’t contain, such as extra copies of a gene or stretches of DNA that were never in the reference.

Phaser, Horizon’s first-of-its-kind assembly-based tertiary analysis module, starts from the assembled genome instead. It reads the two copies of a person’s genome separately and shows what each one holds, gene by gene:

• Extra gene copies: how many copies of a gene a person carries and where they sit, whether side by side or on a different chromosome.
• Unique variants and new DNA: thousands of variants and stretches of sequence, often over 1,000 bases, that lie outside the reference, revealing more disease-causing and disease-associated variants.
• Repeat expansions: stretches of repeated DNA that have grown or shrunk, measured in each copy of the genome.
• Pharmacogenomic alleles: HLA and non-HLA alleles called directly from assemblies, to help assess drug efficacy and adverse reactions.

Some well-known conditions, including Huntington’s disease and CANVAS, are caused by exactly these expansions, and Phaser shows which of the two copies carries them.

In the same pass, Phaser reports the variants on each copy and the person’s HLA types, linked to what they may mean for drug response. On Genome in a Bottle HG002, a standard public benchmark, it found over 99% of duplicated genes with more than 99.9% of its calls correct, and it detects repeats up to 10,000 bases long.

Multimodal: from variant to biology

A variant is a hypothesis; expression is evidence. Horizon^TM 3.0 lets labs upload single-cell transcriptomic and proteomic data, whether from a rare disease patient’s cells or from a tumour, and view it beside reference tissue and single-cell atlases in the same portal.

• Differentially expressed genes in the patient’s or tumour’s data are highlighted against the variants under review.
• Tissue and cell-type atlases show where a gene is normally active, so variants in genes relevant to the affected tissue stand out.
• Protein-level evidence can confirm whether a variant changes what the cell actually makes.

Together with assembly-based analysis, this makes Horizon^TM a single place to go from genome to biology: finding the variant, then showing why it matters.

Built to fit your lab

Horizon^TM 3.0 runs in the cloud, fully on-premises, in-region or hybrid, on CPU or GPU infrastructure. Data is encrypted in transit and at rest, with role-based access control. Horizon^TM is HIPAA-compliant, GenomeArc’s quality management system is certified to ISO 13485:2016, and APIs connect to common LIMS and EHR systems, so Horizon slots into the workflow your lab already run.

See Horizon^TM 3.0 on your own data

Request a live demo with our clinical scientists, the people who built the interpretation engine, and bring a long- or short-read file from a case that is still unsolved.

Request a demo at genomearc.com or email customercare@genomearc.com.