Computational genomics · research and software

Genomic data is abundant. Reliable evidence is harder.

OmicsEdge is an independent computational genomics research and software company. Sequencers generate observations; whether they are evidence depends on what was measurable, where technologies disagree and which benchmark conditions apply. We research those limits and build software that keeps them visible.

Research directions in RNA measurement, variant evidence and benchmarking. Released tool: OmicsEdge Evidence Atlas v1.0.0, 11,553 benchmark events, DOI 10.5281/zenodo.23085673.

SchematicFrom sequencing reads to inspectable evidenceTwo technologies generate reads. Illumina produces many short fragments; Oxford Nanopore produces fewer, much longer reads. The reads are aligned to the same genomic region. At one position, both technologies support the same variant. At another, they disagree: the short reads report a variant that the long reads do not. Context appears. The disputed position sits in a tandem repeat with low mappability, in one sample, judged against one benchmark version. The raw observations are reorganised into measurement, evidence, context, uncertainty and provenance. Evidence should be inspectable.Illumina · short readsOxford Nanopore · long readsreferenceBoth supportDisagreeContexttandem repeatMappabilitysample HG003Illumina · ONTGIAB v4.2.1GRCh38Measurementwhat each workflow observedEvidencedecisions per observationContextwhere in the genomeUncertaintywhat stays unresolvedProvenancewhere each record came fromEvidence should be inspectable.
  1. Two technologies generate reads. Illumina produces many short fragments; Oxford Nanopore produces fewer, much longer reads.
  2. The reads are aligned to the same genomic region.
  3. At one position, both technologies support the same variant.
  4. At another, they disagree: the short reads report a variant that the long reads do not.
  5. Context appears. The disputed position sits in a tandem repeat with low mappability, in one sample, judged against one benchmark version.
  6. The raw observations are reorganised into measurement, evidence, context, uncertainty and provenance.
  7. Evidence should be inspectable.

Who OmicsEdge works with

  • Sequencing and assay teams

    “Which results can we stand behind, and in which regions?”

    Reliability analysis
  • Short- and long-read labs

    “Illumina and ONT disagree here. What does the evidence support?”

    Evidence adjudication
  • Benchmark and method developers

    “Where did this benchmark result come from, and does it apply to us?”

    The Evidence Atlas
  • Work with us

    Research collaborations and applied analysis on public or your own data.

    Start a conversation

The problem

Different technologies observe biology differently.

Six steps from raw sequencing output to an answer you can defend. Scroll through them.

  1. 01 · Raw data

    Sequencing generates observations.

    Short fragments, long reads, raw signal. Producing them is now routine, and there are a great many of them.

  2. 02 · Measurement

    Not everything is equally measurable.

    Observations pile up unevenly. Some regions and molecules are covered deeply; others barely at all. What a workflow could measure is the first boundary on what it can tell you.

  3. 03 · Disagreement

    Technologies can disagree.

    Illumina and Oxford Nanopore can look at the same sample and report different things at the same position. Most calls agree. The ones that do not are where the questions are.

  4. 04 · Context

    Context changes what a call means.

    Homopolymers, tandem repeats, low-mappability and duplicated sequence are where methods are known to behave differently, which is why benchmarks stratify by them. A call should not be read without knowing where it was made.

  5. 05 · Evidence

    Observations become evidence when their conditions stay attached.

    Sample, platform, benchmark version, comparator and source file travel with each record. Exact relationships are drawn solid; unresolved ones stay dashed and visible.

  6. 06 · Decision

    An honest answer can be “not enough evidence”.

    Supported, discordant, unresolved, or abstain. Sometimes the correct computational output is not another confident prediction.

When technologies disagree

Sometimes the correct answer is: there is not enough evidence.

A standard workflow collapses disagreement into PASS or FAIL. We are exploring evidence adjudication instead: given everything Illumina and Oxford Nanopore report about a candidate variant, which interpretation can be supported, and when should a system abstain?

Illustrative · not model outputPaired Illumina and Oxford Nanopore evidence converging on a candidate variant, weighed, and resolved to a support state or ABSTAINIllumina readsONT readscontext: non-difficultcontext: homopolymerreads cannot be placed uniquelycontext: low mappabilitycontext: segmental duplicationcontext: outside benchmark regionscandidatevariantEvidence weighedIllumina call evidenceONT call evidenceQuality consistencyContext difficultyInterpretationBOTH_SUPPORTEDILLUMINA_SUPPORTEDONT_SUPPORTEDNEITHER_SUPPORTEDABSTAINUNASSESSABLEbenchmark labelling only

Both platforms call the same allele with strong, consistent evidence in easy sequence. The supportable interpretation is that both calls hold.

Illustrative cases showing what each state means. They are not predictions: the adjudication model has not been evaluated and no performance figures exist. Evidence levels are drawn qualitatively on purpose; the eventual system reads 42 approved features.
  • Clear agreement: BOTH_SUPPORTED. Both platforms call the same allele with strong, consistent evidence in easy sequence. The supportable interpretation is that both calls hold.
  • Illumina-only indel in a homopolymer: ILLUMINA_SUPPORTED. Short reads call a small deletion in a run of one base; long reads do not. Context matters: homopolymers are a known difficulty for some long-read chemistries.
  • ONT-only call in low mappability: ONT_SUPPORTED. Long reads span a region where short reads cannot be placed uniquely. The absence of a short-read call is weak evidence here.
  • Conflicting evidence: ABSTAIN. Both platforms call something, with different genotypes, low quality and in duplicated sequence. The honest output is: not enough evidence for an automated conclusion.
  • Benchmark cannot judge: UNASSESSABLE. The truth set cannot establish whether either call is right. This is benchmark uncertainty, not model abstention: such cases are excluded from training labels.

What we are building

Three independent directions, one idea: make the boundaries of evidence visible.

Each is a separate research programme with its own data and methods, at its own stage: one completed study, one method in development, one released tool. None depends on another.

01 · Research

RNA Measurement Intelligence Research

RNA-sequencing does not observe every transcript equally. Abundance, sequence, architecture, isoform similarity and the workflow itself decide what can be seen. We are building toward a measurement-intelligence layer that would tell a researcher, before or after an experiment, which transcripts a given workflow is likely to measure reliably.

Could support: experimental design, assay selection, transcript-level QC, workflow comparison.

How it could work

Transcript

Properties

IllustrativeHow transcript properties and sequencing workflow shape what can be observedTranscriptIllumina short-readNanopore direct-RNAObservationabundantE1E2E3E4E5long transcriptstrongstrongclassBOTHrareE1E2E3E4E5long transcriptno readsweakmissingclassILLUMINA_ONLYmoderateE1E2E3E4E5long transcriptisoform B shares these exonsuncertainstrongclassDIRECT_RNA_ONLYmoderateE1E2short transcriptlow mappability: sequence repeated elsewhereno readsuncertainmissingclassINDETERMINATE

Plenty of molecules and unique sequence: both workflows see it clearly.

Structure is drawn because it was tested. In the K562 study, transcript-wide measured in-vivo structure added negligible predictive value beyond the other properties shown.

Illustrative, not study data. Outcome classes use the study’s vocabulary (BOTH, ILLUMINA_ONLY, DIRECT_RNA_ONLY, NEITHER, INDETERMINATE). Real observability depends on the specific protocol, depth, quantifier and thresholds. Modelling that dependence explicitly is the research direction.
Illustrative outcomes by transcript type
TranscriptIllumina short-readNanopore direct-RNAClassWhy
Abundant, distinctstrongstrongBOTHPlenty of molecules and unique sequence: both workflows see it clearly.
RareweakmissingILLUMINA_ONLYFew molecules. Deep short-read sampling catches a little; a lower-throughput direct-RNA run may catch nothing.
Shares exons with an isoformuncertainstrongDIRECT_RNA_ONLYShort reads on shared exons cannot be assigned to one isoform. Full-length reads can tell the isoforms apart.
Short, repetitive sequenceuncertainmissingINDETERMINATESequence that reads cannot be placed uniquely, and too short to anchor long reads: neither workflow gives a clean answer.

02 · In development

Variant Evidence Adjudication In development

When Illumina and Oxford Nanopore disagree about a candidate variant, the useful output is not PASS or FAIL. It is a supportable interpretation, or an explicit abstention. Protocols and 42 approved features are frozen. No model results exist yet, and we will not show any until they do.

The concept
  • BOTH_SUPPORTED
  • ILLUMINA_SUPPORTED
  • ONT_SUPPORTED
  • NEITHER_SUPPORTED
  • ABSTAIN not enough evidence

03 · Released

OmicsEdge Evidence Atlas Released v1.0.0

Benchmark results usually arrive as static files and aggregate accuracy. The Evidence Atlas keeps them as evidence instead: each run-scoped event, both of its observations, and the sample, platform, pipeline, comparator and artifacts that produced them. It is a provenance-rich evidence resource, not a predictive model.

v1.0.0 · 11,553 events · 23,106 observations · 4,278 variants · DOI 10.5281/zenodo.23085673

  1. Eventchr8:3,989,216 T>Crun-scoped record · 1-based GRCh38
  2. Sample and platformHG003 · Illumina NovaSeq 6000coverage 35x
  3. PipelineRN-IlluminaNeuSomatic, GATK 4.1.7.0, Picard 2.22.4
  4. Benchmark conditionshap.py 0.3.15 · vcfeval 3.12.1truth artifact-truth-hg003-grch38-v421
Observations (two sides)truth TP · query TPGT 0/1 · match gm · recorded comparator outcome
Provenanceartifact-query-60z59-hg003 → run-phase2c-hg003-60z59-grch38-v421obs-run-phase2c-hg003-60z59-grch38-v421-1226-query
One real record from the Evidence Atlas. Not a universal trust score: evidence you can inspect, condition by condition.

Research foundation

Each direction rests on frozen, public-data research.

  1. What determines whether RNA can be reliably observed?

    A locked K562 study tested whether measured in-vivo RNA structure adds predictive information beyond abundance, sequence, architecture, identifiability and callability. Within that setting, it added negligible value. A negative result, reported as one.

    Manuscript submitted NAR Genomics and Bioinformatics

    The study
  2. How should software reason about sequencing disagreement?

    Paired Illumina and Oxford Nanopore evidence, frozen preprocessing and baseline protocols, deterministic extraction of 42 approved features, and selective prediction with abstention.

    In development no model results yet

    The protocol
  3. How can benchmark evidence stay inspectable?

    The OmicsEdge Evidence Atlas: a provenance-preserving record of 11,553 benchmark events across HG002, HG003 and HG004, Illumina and Oxford Nanopore. Released as v1.0.0.

    Released v1.0.0 DOI 10.5281/zenodo.23085673

    The release

A released OmicsEdge tool Live

Inspect real benchmark evidence, now.

The OmicsEdge Evidence Atlas is a released OmicsEdge research resource, and a concrete example of how we work: a real problem, a frozen protocol, public data and a usable tool. Its Evidence Explorer searches all 11,553 events in your browser, by locus, allele, sample, technology or identifier. Opening an event loads only its evidence shard, with both observations and their provenance.

Retrieval only. Nothing is scored, ranked or merged.

Open the Evidence Explorer

Example from the release · chr8:3,989,216 T>C

variant:sha256:00e903d4b12125d074bb62200…

Linked benchmark events for this variant, with recorded truth-side and query-side decisions
SampleTechnologyTruth sideQuery sideGTOpen
HG003IlluminaTPTP0/1Open
HG003Oxford NanoporeTPTP0/1Open
HG004IlluminaTPTP1/1Open
HG004Oxford NanoporeTPTP1/1Open

4 run-scoped events linked by exact normalized-allele identity: 8 observations. Recorded comparator outcomes, shown side by side; not merged or ranked.

Open in the Evidence Explorer

Benchmark evidence is conditional

A benchmark outcome is only true of its conditions.

The same variant can be benchmarked under different samples, call sets, pipelines, truth sets and benchmark regions. These are the six benchmark runs in the released Evidence Atlas, read directly from its frozen data. Every recorded TP, FP or FN in the Atlas carries one of these sets of conditions with it.

Benchmark runs in the OmicsEdge Evidence Atlas v1 and their recorded conditions
SampleTechnologyPlatformCoveragePipeline and callersTruth setBenchmark regionsEvents
HG002IlluminaNovaSeq 600035xRN-Illumina
NeuSomatic, GATK 4.1.7.0, Picard 2.22.4
artifact-truth-hg002-grch38-v421artifact-bed-hg002-grch38-v4211,804
HG002Oxford NanoporePromethION R9.447xPEPPER-DeepVariant (ONT)
PEPPER, DeepVariant
artifact-truth-hg002-grch38-v421artifact-bed-hg002-grch38-v4211,798
HG003IlluminaNovaSeq 600035xRN-Illumina
NeuSomatic, GATK 4.1.7.0, Picard 2.22.4
artifact-truth-hg003-grch38-v421artifact-phase2-frozen-panel1,850
HG003Oxford NanoporePromethION R9.485xPEPPER-DeepVariant (ONT)
PEPPER, DeepVariant
artifact-truth-hg003-grch38-v421artifact-phase2-frozen-panel1,858
HG004IlluminaNovaSeq 600035xRN-Illumina
NeuSomatic, GATK 4.1.7.0, Picard 2.22.4
artifact-truth-hg004-grch38-v421artifact-phase2-frozen-panel2,117
HG004Oxford NanoporePromethION R9.485xPEPPER-DeepVariant (ONT)
PEPPER, DeepVariant
artifact-truth-hg004-grch38-v421artifact-phase2-frozen-panel2,126
Every recorded outcome inherits these conditions. All runs used hap.py 0.3.15 with vcfeval 3.12.1 on GRCh38. The same TP or FN means something specific to its sample, call set, pipeline, truth set and benchmark regions. The Atlas keeps them attached, and does not total outcomes into a ranking of platforms.

Released tool · OmicsEdge Evidence Atlas v1.0.0

What a released OmicsEdge resource looks like.

Counts read from the frozen release manifest and search index. They describe the research dataset, not usage or business metrics. HG002 comes from the Phase 1 proof of concept on chr20; HG003 and HG004 from the Phase 2 panel.

11,553benchmark eventsrun-scoped
23,106observationstruth and query side
4,278biological variantsexact normalized-allele identity
3samplesHG002 · HG003 · HG004
2sequencing technologiesIllumina · Oxford Nanopore
20chromosomesGRCh38
60unresolved eventskept visible, never guessed
256evidence shardsloaded one at a time

Data lineage

Every record knows where it came from.

Follow one Evidence Atlas record from the public benchmark release down to the biological variant it links to. Scroll: the chain assembles as you go.

Provenance is what lets you tell apart the same biological variant, the same benchmark event, the same comparator decision and the same technology behaviour. The Atlas never assumes they are equivalent.

  1. 01 · Source benchmark

    Truth set artifact-truth-hg003-grch38-v421

    • query call set: artifact-query-60z59-hg003
    • benchmark regions: artifact-phase2-frozen-panel
  2. 02 · Version

    hap.py 0.3.15 · vcfeval 3.12.1

    • runtime image d63b963a6cb0…
    • reference: artifact-reference-grch38-no-alt
  3. 03 · Sample

    HG003

    • experiment exp-pfda-v2-60z59-hg003
  4. 04 · Platform

    Illumina NovaSeq 6000

    • pipeline RN-Illumina · 35x
  5. 05 · Event

    chr8:3,989,216 T>C (1-based, GRCh38)

    • event-run-phase2c-hg003-60z59-grch38-v421-chr8-record-1226
    • benchmark run run-phase2c-hg003-60z59-grch38-v421
  6. 06 · Observations

    truth TP · query TP · GT 0/1 · match gm

    • obs-run-phase2c-hg003-60z59-grch38-v421-1226-truth
    • obs-run-phase2c-hg003-60z59-grch38-v421-1226-query
  7. 07 · Atlas record

    EXACT_NORMALIZED_ALLELE → biological variant

    • event-variant-link:sha256:7bb5…
    • variant:sha256:00e903d4b12125d074b…
    • 4 linked events across 2 samples and 2 technologies
A real record, end to end. The variant the frozen release selects in its own deterministic product query. Every line is read from the Evidence Atlas v1 bundle. Open it in the Evidence Explorer.

Applied work

The same discipline, applied to your data.

Selectively, we help teams design benchmarks, validate workflows and analyse sequencing data with explicit evaluation boundaries and preserved provenance.

Applied capabilities
  • Benchmark design and validation
  • Workflow and pipeline validation
  • Long-read and short-read variant analysis
  • RNA-seq measurement analysis
  • Reproducible pipelines, HPC and cloud
  • Scientific visualization

Evidence should be inspectable.

Tell us about the measurement, benchmarking or analysis problem you are working on: research collaborations and applied computational genomics work. Or look at a released tool first.