The interpretation layer

Sequencing is solved.
Interpretation is the answer.

Soma Nova turns raw genome into a defensible clinical result — diagnoses, drug-response guidance, and risk — with a qualified human signing off every call.

Every clinical assertion is evidence-linked and human-reviewed.

Standards & complianceHL7 FHIRGA4GHACMG/AMPCPICGDPRISO 27001
The bottleneck

The lab is fast. The desk is the constraint.

A single genome yields millions of variants and a handful that matter. Resolving them still depends on scarce specialists and manual curation that does not scale.

3.2B
base pairs in a single human genome
4–5M
variants called against the reference
<10
typically meaningful for the clinical question

Sequencing is commoditising. Interpretation is the durable, defensible layer.

The platform · GIP

One platform, understood in layers.

The Genomics Intelligence Platform is built as distinct layers — how it reasons, how it runs, and how it accumulates a lifelong model of the patient. Each is shown below.

Layer view 01 — Intelligence

A reasoning system, not a black box.

Learned models are grounded by an explicit, curated knowledge graph. Reading top-down: copilots serve the clinician; every layer beneath exists to make the layer above defensible.

InterfaceAI copilots
Role-aware assistants for the geneticist, physician and researcher — the surface a human actually works in.
ReasoningReasoning engine · XAI
Applies ACMG/AMP-style guideline logic transparently and attaches evidence, sources and confidence to every assertion.
LanguageMedical LLM + RAG
Drafts and synthesises over retrieved, cited evidence — never from parametric memory. No citation, no assertion.
EvidenceKnowledge graph · GNNs
Curated genes, variants, diseases and drugs, with graph neural networks scoring gene–disease and variant relationships.
FoundationTransformer & foundation models
Sequence-aware models for variant effect and phenotype matching — the learned substrate everything above is grounded against.
↓ learned & generalgrounded & explainable ↑
Layer view 02 — Architecture

Cloud-native, standards-based, API-first.

Stateless interpretation scales horizontally; a shared, stateful evidence substrate keeps every result consistent and reproducible years later.

ClientsEHR Lab systemsAPI partners
│ SMART on FHIR ▼
API Gateway — auth · rate limiting · routing · marketplace
Ingestion
FASTQ · BAM/CRAM · VCF
Interpretation
annotate · prioritise · reason
Reporting
HL7 FHIR Genomics
Digital Twin
longitudinal model
Event streaming · Data lakehouse · Vector DB · Knowledge graph
Kubernetes · containerised · MeitY-empanelled / regional & sovereign cloud
Cloud-nativeCompute scales with volume; no over-provisioning.
MicroservicesIndependently deployable, independently scaled.
API-firstEvery capability governed behind the gateway.
Standards-nativeFHIR & GA4GH built in, not adapted.
Layer view 03 — The Digital Twin

A living, computable model of the patient.

The genome is fixed; its meaning is re-read as knowledge, phenotype and environment evolve. Nine data layers fuse into one lifelong asset.

01

Genome twin

WGS variants & annotations

Stable genetic baseline, reinterpretable over a lifetime.

02

Clinical twin

diagnoses · history · meds

Context for genotype–phenotype reasoning.

03

Phenotype twin

HPO-coded signs

Drives phenotype-aware prioritisation.

04

Radiology twin

imaging findings

Correlates structural with molecular signal.

05

Laboratory twin

labs & biomarkers

Dynamic physiological state.

06

Lifestyle twin

behaviour · diet · activity

Modifiable risk factors.

07

Wearable twin

continuous device signals

Real-time monitoring and trends.

08

Environmental twin

exposures · geography

Gene–environment interaction.

09

Population twin

cohort context

Comparative and public-health inference.

Why it holds up

Built for software that lives next to patient care.

Explainability

Every assertion carries its evidence.

No ungrounded output. Each call links to the criteria and citations behind it — auditable years later, not just today.

Human in the loop

AI augments the expert; it never replaces them.

A criteria engine classifies, an LLM synthesises, and a named clinical reviewer stands behind every result.

Standards-native

Built on FHIR and GA4GH, not around them.

Interoperability is the strategy. Reports drop into existing LIMS and EHR workflows without a translation layer.

Sovereign by design

Cloud-native, with data residency where required.

Cloud-agnostic Kubernetes lets health systems and governments keep data inside their own borders.

Solutions

One interpretation layer. Six clinical questions.

From the germline to the whole population — each module built for the buyer who owns the decision.

Trust & compliance

Genomic data demands the highest bar.

Permanent, familial and predictive — a breach affects relatives who never consented. Security is core architecture, engineered in from the start.

Interop
HL7 FHIR · GA4GH

Native standards for genomic exchange and reporting.

Data
GDPR · DPDP · Residency

Zero-trust, encryption everywhere, data kept in-region.

Classification
ACMG/AMP · CPIC

Transparent criteria engines, not opaque black boxes.

Assurance
ISO 27001 · SOC 2

Immutable audit trail and versioned knowledge per case.

Anchor sites

Prove it on your own cases.

Anchor sites are partners, not customers. Commit a defined case volume and named reviewers; get early capability and a real say in the roadmap.

Rare disease · Oncology · Population — India & UK first.