Semantic Identity Will Decide Whether the Novartis Abogen RNA Partnership Scales

by Marina VasquezHead of Semantic Systems, HMND

Novartis and Abogen have created a concrete test for cross company scientific identity. Their agreement gives Novartis exclusive worldwide rights to ABO2203 and options on other RNA programs, with Abogen eligible for a reported $575 million upfront payment and up to approximately $7.2 billion in conditional milestones. The scientific opportunity is large. The data boundary is larger.

ABO2203 is an mRNA encoded T cell engager designed to direct activity toward CD3 and CD19, with the aim of depleting disease causing B cells. That mechanism makes the handoff unusually demanding: the partners must preserve identity across sequence design, delivery system, potency testing, manufacturing, clinical material, and the evidence connecting each one.

A multinational RNA partnership succeeds when the same candidate can be found, understood, trusted, and acted upon across laboratories, companies, jurisdictions, and stages of development. That candidate needs a stable identity as it moves from construct design to formulation, from assay to sample, from batch to study, and from study to submission. Every handoff tests the meaning of the entity.

Many partnerships quietly fail in the first question nobody can answer without opening six systems.

The candidate that changes names

ABO2203 may begin inside Abogen as a program identifier connected to a particular mRNA construct, encoded target, delivery system, and development hypothesis. Novartis will place that program inside its own systems, governance, clinical vocabulary, manufacturing records, and regulatory plans. The two organizations can exchange files for years while describing different objects with the same label.

The entity that should exist is a governed candidate record with durable identifiers for the program, molecular construct, sequence version, formulation, intended use, and legal owner. A display name can change. The identity cannot drift with the spreadsheet that happens to contain it.

A construct differs from a candidate. A candidate differs from a batch. A batch differs from a sample. A sample differs from an assay result. Those distinctions sound elementary until a translational team asks whether a potency result belongs to the current sequence, the previous formulation, or a retained sample prepared under a different manufacturing process.

Here is the small correction that tends to arrive after the expensive meeting: a platform does not constitute an entity. “Abogen RNA platform” describes a family of capabilities and assets. It does not identify the molecule, process, batch, or evidence supporting a decision.

Meaning has an owner

Ontology ownership determines who can define the terms that cross the partnership boundary. If Abogen owns the construct vocabulary and Novartis owns the clinical study vocabulary, both sides need a shared contract for the terms that connect them. A local ontology can remain local. The bridge between ontologies requires named custodians, change rules, mappings, and effective dates.

The entity that should exist is a partnership semantic model with explicit ownership for each class and relationship. Someone must decide whether “same formulation” means identical composition, equivalent composition within an approved tolerance, or a formulation family with a documented change history. Someone must decide whether “sample” means a physical aliquot, a parent specimen, or a laboratory accession.

Teams often postpone these decisions because the initial exchange feels urgent. The delay stores ambiguity in every downstream record. Later, a migration project gets called integration, although the work consists of asking scientists what their nouns meant three years ago.

I have watched this happen around assay data. A receiving team sees a potency value, a unit, and a date. The originating team knows the cell line passage, instrument method, reference standard, analyst, acceptance range, and protocol revision. Both teams possess the result. Only one possesses its provenance.

A graph is a chain of accountable facts

A useful knowledge graph would connect the candidate to its construct, the construct to its sequence version, the sequence to its production batch, the batch to released material, the material to a sample, the sample to an assay execution, and the assay to a study decision. Each edge needs a source, timestamp, responsible system, and status.

Provenance gives the graph its memory. Versioning gives it time.

A sequence update should create a new version with a reason, approval state, and relationship to the predecessor. A revised assay method should preserve the historical method used to generate earlier results. A manufacturing change should connect the old and new process descriptions without collapsing them into one convenient “current” record.

Entity resolution then becomes a scientific operation. The system must determine that two records refer to the same physical batch or the same logical candidate, while preserving uncertainty when the evidence is incomplete. A fuzzy name match cannot make that decision. It can suggest a relationship for review. Sequence fingerprints, material identifiers, controlled attributes, source system references, and signed transfer records can support it.

Federated query versus export archaeology

A federated query asks several governed systems for related facts while preserving their source context. It can return the candidate record from one boundary, the batch record from another, and the study result from a third, then show how the links were established and which permissions shaped the answer.

A pile of exports produces files.

The distinction matters when a Novartis scientist asks whether a clinical result came from material manufactured under the process now proposed for a regulatory package. A federated system can traverse the relationship and expose the relevant versions, approvals, and provenance. Export archaeology leaves the scientist comparing filenames, timestamps, and institutional memory.

Security makes the problem more exact. Abogen may retain source data in one jurisdiction while Novartis operates a regulated development environment in another. A semantic layer can expose approved metadata and relationship evidence without moving every restricted record across the boundary. Access control must apply to entities and relationships, not merely folders. A user may be allowed to know that a sample exists while being prohibited from viewing its donor level data.

That is an infrastructure decision. The architecture may involve APIs, graph storage, federated identity, metadata catalogs, policy enforcement, and audit logs. None of those components repairs an undefined sample.

Decisions change when identity travels

Translational researchers can compare results across assays when the system preserves the relationship between a result, its sample, its material state, and its method version. That changes which signals deserve follow up. A weak result from an obsolete construct should not carry the same weight as a weak result from the clinical candidate.

Clinical development teams can connect protocol versions, investigational product versions, sites, samples, and analyses without treating every data cut as a new universe. The study record gains a traceable relationship to the material actually administered. Data review becomes faster because the question can be answered through identity rather than reconstructed through correspondence.

CMC teams can assess comparability with a chain that joins process parameters, analytical methods, batch disposition, deviations, and stability observations. A formulation change becomes a visible event with consequences, rather than a renamed column in a transfer workbook.

Regulatory preparation gains the same discipline. Submission content can point back to controlled source records and show which version supported each claim. The resulting package carries an evidentiary structure that can survive review, query, and later variation.

The regulator does not need a fashionable graph. The regulator needs a defensible answer to a plain question: which material, produced how, tested by which method, supported this conclusion?

The handoff is the product

Partnership leaders often measure transfer by delivered documents, completed milestones, and populated repositories. Those measures miss the failure that appears when a scientist cannot establish whether two records describe the same thing.

The handoff should instead produce a semantic agreement. It should define shared identifiers, ownership, mappings, provenance requirements, version behavior, access boundaries, and escalation paths for ambiguous records. The agreement belongs beside the quality agreement because it governs the meaning of the evidence that quality decisions consume.

Teams will stall when ontology work is treated as documentation after the software build. Engineers then encode accidental definitions into schemas and interfaces. Scientists compensate with side files. Analysts compensate with manual joins. Security teams block broad access because nobody can explain which relationship exposes which protected fact. Everyone sighs at another silo labeled integration.

Novartis and Abogen now have a chance to make the partnership’s data boundary an operating asset. ABO2203 can retain one identity while its descriptions, measurements, manufacturing records, and clinical evidence accumulate across organizations. That requires owners willing to settle meanings before the next transfer, and systems designed to preserve those decisions.

A dashboard can display the confusion beautifully. A semantic system can make the confusion answerable.

If this handoff problem is sitting inside your research stack, hello@example.com is a sensible place to put the first hard question.