Interactional ontology

Draft. This monograph is an unreviewed draft. Its sources have not been checked by a named person and no domain reviewer has approved it. Treat every claim as provisional.

Definition and scope

Interactional ontology is the NucLex proposal to represent interactions as explicit knowledge objects: records that name the participating entities, the direction of effect where one exists, the biological and experimental context, the quantitative parameters with their units, the uncertainty attached to those parameters, the evidence behind them, the provenance of the record itself, and the version under which all of this was asserted. The phrase comes from the founder's own formulation in the NucLex discussion of 19 August 2026, where the requirement moved from "the property of molecule A" to "how molecule A behaves to molecule B" (discussion record, section 8). It is a design direction, not an implemented standard, and this monograph is written so that a reader can tell at every point which statements describe NucLex's proposal and which describe representations that already exist and are in wide use.

The scope is deliberately narrow. The monograph does not propose a complete schema, does not claim that NucLex holds any curated interaction records, and does not provide reusable parameter values. Every example below is synthetic and labeled as such. The purpose is to answer the central question, how an interaction can be a knowledge object, precisely enough that a later engineering specification could be tested against it.

Key distinctions

The first distinction is between three kinds of knowledge object that are routinely conflated.

Properties of molecule A are statements true of A regardless of partner: its chemical identity, its molecular mass, the radionuclide it carries and that radionuclide's half-life, its charge at physiological pH, its structure. These belong in an entity record. A Radiopharmaceutical entry and a Radionuclide entry are of this kind.

Properties of molecule B are, likewise, statements true of B alone: a membrane protein's amino-acid sequence, its subcellular location, the cell types in which it is expressed and at what level, its known biological roles. A Molecular target entry is of this kind, and the guidance for that entry already warns that identity, expression, accessibility, and biological role need separate evidence.

A interacting with B is a third thing. "A binds B with a dissociation constant of so many nanomolar" is not a property of A and not a property of B. It is a property of the pair, measured under conditions that belong to neither molecule: a buffer, a temperature, a cell line or a purified protein, a radiolabel that may itself alter binding, an assay format that reports an inhibition constant rather than an equilibrium constant. Change the partner and the value is meaningless; change the conditions and the value may change by an order of magnitude. The interaction is therefore a distinct Interaction object with its own identity, its own evidence, and its own version history.

The second distinction is between a qualitative assertion and a quantitative Parameter. "A binds B" is a relationship with a type and a direction. "A binds B with an association rate constant of so many per molar per second at 37 degrees Celsius" is a parameter attached to that relationship. Established knowledge graphs often record the first; computational models need the second.

The third distinction is between an interaction record and a model. A record states what is known, with its Uncertainty and Provenance. A model selects records, makes further assumptions, and computes. The monograph on Model-ready knowledge exchange addresses what must travel with a parameter when it crosses from record to model; the monograph on Theranostic digital twins addresses what a model built from such records would need to represent and validate.

Historical development

The idea that interactions deserve records of their own is not new, and honesty about this is the review gate for the monograph. In molecular biology, the growth of high-throughput protein interaction data in the late 1990s and early 2000s created a need for a shared exchange format, and the Human Proteome Organization's Proteomics Standards Initiative published the PSI-MI (Proteomics Standards Initiative Molecular Interaction) format in 2004, extended to level 2.5 in 2007 (Hermjakob et al., 2004; Kerrien et al., 2007). In systems biology, the Systems Biology Markup Language (SBML), first described in 2003, gave reactions, species, compartments, and kinetic parameters a machine-readable form so that models could move between simulation tools (Hucka et al., 2003). BioPAX, published as a community standard in 2010, represented pathways, including molecular interactions and their participants, in an ontology expressed in the Web Ontology Language (OWL) (Demir et al., 2010). The Gene Ontology, from 2000, took a different route: its molecular function branch classifies the activities a gene product can perform, including binding, as terms in a controlled hierarchy rather than as records of particular pairwise events (Ashburner et al., 2000).

Two further developments matter for NucLex. The MIRIAM guidelines of 2005 stated the minimum annotation a model needs to be identifiable, reproducible, and connected to external resources (Le Novère et al., 2005), and the COMBINE archive, with its OMEX (Open Modeling EXchange) manifest format, packaged a model with its data, simulation description, and metadata in one file (Bergmann et al., 2014). The World Wide Web Consortium's PROV-O ontology, a recommendation from 2013, gave provenance a standard vocabulary of entities, activities, and agents (W3C, 2013). The question for NucLex is therefore not whether interactions can be represented. They can, and have been for two decades. The question is what representation a nuclear medicine physician, radiochemist, or modeler needs that these standards do not deliver together, in one place, with the evidence and uncertainty attached.

Philosophical or technical account

An interaction record, in the NucLex proposal, would need the following components. Each is stated as a requirement with the reason it is needed.

Participants. At least two entity references, each resolving to an entity record with its own identity. The record must say which role each participant plays (ligand, target, enzyme, substrate, transporter, cargo). For a radiopharmaceutical the ligand record must identify the labeled construct actually studied, because a cold precursor, a metal-chelated analogue, and the radiolabeled product can bind differently.

Direction. Some interactions are symmetric in form but asymmetric in consequence: binding is mutual, but internalization is something the cell does to the ligand. The record should state direction where it is meaningful and state explicitly that it is not meaningful otherwise, rather than leaving an unlabeled edge.

Interaction type. A term from a stated vocabulary: reversible binding, covalent modification, transport, enzymatic conversion, competitive inhibition. The vocabulary should be named and versioned so that a later reader knows which classification was in force.

Context. The conditions under which the assertion holds: biological system (purified protein, cell line with its identifier and source, tissue, animal species and strain, human), Cell type where relevant, temperature, pH, buffer or medium, presence of competitors, time after administration where kinetics matter. The founder's requirement was that NucLex record "properties and relations" that a model can take as inputs; without context those relations cannot be matched to the model's own assumptions. The Context entry explains why biological setting, method, time, and intended use affect applicability.

Parameters with units. Each quantitative claim is a parameter with a name, a value, a unit, and a parameter type (equilibrium dissociation constant, inhibition constant, association rate, dissociation rate, internalization half-time, maximum binding capacity). Units are not optional. A value without a unit is not a parameter; it is a number.

Uncertainty. A parameter carries the uncertainty the source reported, in the form the source used: standard deviation, standard error, confidence interval, range across replicates, or an explicit statement that no uncertainty was reported. The record must not synthesize a single confidence number from heterogeneous sources. The Uncertainty entry gives the reason: measurement uncertainty, estimation uncertainty, and applicability uncertainty are different quantities and collapsing them hides which one dominates.

Evidence. A link to each source that supports the parameter, with the location in the source, the experimental method, and the evidence type. The Evidence entry distinguishes strength from relevance; a precise measurement in the wrong system may be less useful than a rougher one in the right system.

Provenance. Who or what created the record, from which sources, by which transformation (manual curation, assisted extraction with a named model and prompt version, import from an external database with its release), and when. Provenance of the record is distinct from evidence for the claim. The Provenance entry and the monograph on Human review of machine proposals explain the principle that an assisted extraction remains a proposal until a named person approves it.

Version. The record's own version, the version of the vocabulary used for interaction types, and the versions of the entity records it references. A parameter retrieved by a model must be retrievable again, identically, from the same version.

Status. Whether the record is a candidate, reviewed, or deprecated, and by whom. This is the hinge between the interactional layer and the governance workflow.

The following is a synthetic teaching example. Every identifier is a local NucLex teaching identifier, every value is invented, and nothing in it may be reused as a parameter.

# SYNTHETIC EXAMPLE. Local teaching identifiers only (NXT- prefix).
# All values are invented for illustration and are not measurements.
record_type: interaction
id: NXT-I-0001
version: "0.1.0"
status: candidate
interaction_type:
  label: reversible non-covalent binding
  vocabulary: NucLex interaction type list (teaching draft), version 0
participants:
  - entity: NXT-E-0101
    label: "radiolabeled small-molecule ligand (synthetic)"
    role: ligand
    note: "The labeled construct, not the cold precursor."
  - entity: NXT-E-0202
    label: "transmembrane glycoprotein target (synthetic)"
    role: target
direction:
  stated: false
  note: "Binding is mutual; internalization is recorded as NXT-I-0002."
context:
  system: "cultured human cell line expressing the target (synthetic)"
  cell_line_identifier: "NXT-CL-0003"
  species: "Homo sapiens"
  temperature:
    value: 4
    unit: "degree Celsius"
  assay: "saturation binding, membrane preparation"
  competitors: none
  time_after_addition:
    value: 60
    unit: minute
parameters:
  - name: equilibrium_dissociation_constant
    symbol: Kd
    value: 8.0
    unit: nanomolar
    uncertainty:
      type: standard deviation
      value: 2.5
      replicates: 3
  - name: maximum_binding_capacity
    symbol: Bmax
    value: 120000
    unit: "sites per cell"
    uncertainty:
      type: not reported
evidence:
  - source: NXT-S-0500
    description: "synthetic in vitro report used for teaching"
    location: "Table 2"
    method: "radioligand saturation assay"
    evidence_type: "direct experimental measurement"
provenance:
  created_by: "NucLex editorial team (teaching draft)"
  method: "manual authoring of a synthetic example"
  created: 2026-10-10
  assisted_extraction: null
review:
  state: "not reviewed"
  reviewer: ""
applicability_note: >-
  Measured at 4 degrees Celsius in a membrane preparation. Use at 37 degrees Celsius
  in a living cell requires a stated assumption and separate evidence.

The example shows the shape, not the content. The field that most distinguishes the proposal from a plain data table is the applicability note together with the explicit context block: the record carries the conditions that limit its own reuse.

Biomedical relevance

Established representations already cover large parts of this territory, and the reader should be able to see exactly where the NucLex proposal would overlap, reuse, or diverge.

BioPAX represents pathways as an OWL ontology with classes for physical entities, interactions, and their participants, including molecular interactions, biochemical reactions, and control relationships (Demir et al., 2010). It can carry evidence and cross-references. It was designed for pathway databases and exchange, with a qualitative emphasis; quantitative kinetic parameters are not its center of gravity. NucLex would have no reason to reinvent the participant and interaction classes BioPAX provides, and a reviewer should ask whether any proposed NucLex interaction type lacks a BioPAX counterpart.

SBML represents a model, not a knowledge base: species, compartments, reactions with kinetic laws, parameters with units, and events (Hucka et al., 2003; SBML, What is SBML). An SBML reaction is the executable form of an interaction, stripped to what a simulator needs. It carries annotations that can point to external resources, which is where MIRIAM applies. SBML does not itself store the evidence, the assay conditions, or the uncertainty of a parameter in a structured way; those belong in annotation or in accompanying documents. The NucLex proposal is, in this light, a candidate upstream source for SBML parameters, with the context and uncertainty that SBML expects the modeler to have handled before export.

The Gene Ontology's molecular function branch classifies what a gene product can do (Ashburner et al., 2000). A term such as a receptor binding activity says that products of a gene are annotated with that capacity, supported by an evidence code. It does not record a particular pair under particular conditions with a particular constant. GO answers "what kind of activity" and the interactional record answers "this pair, these conditions, this value."

PSI-MI was built precisely to exchange molecular interaction records with participants, detection methods, experimental context, and controlled vocabularies for interaction types and evidence (Hermjakob et al., 2004; Kerrien et al., 2007). It is the closest established analogue to what the NucLex proposal describes, and the honest statement is that most of the structural requirements listed above are already expressible in PSI-MI 2.5. The gap, if there is one, lies in quantitative parameters with uncertainty and in radiopharmaceutical-specific context: the identity of the labeled construct, the radionuclide and specific activity at the time of the assay, and the relation of in vitro binding to in vivo retention. Whether that gap justifies a new design or an extension of PSI-MI is an open question that the review process must answer, not this monograph.

Reactome is a curated pathway knowledge base that represents reactions with inputs, outputs, catalysts, and literature references, organized into pathways and expressed in a schema that can be exported to BioPAX and SBML (Jassal et al., 2020). It is an example of what disciplined human curation of interactions looks like at scale, and its editorial model, with authors and reviewers named on each reaction, is closer to what NucLex intends than any file format is.

STRING is a database of protein-protein associations drawn from experiments, curated databases, text mining, and computational prediction, each association carrying a combined score (Szklarczyk et al., 2019). It is valuable for discovery and is the clearest counterexample to the NucLex proposal in one respect: its combined score blends heterogeneous evidence into a single number, which is exactly what the Uncertainty scope instruction asks NucLex not to do for parameters intended as model inputs. STRING is right for its purpose; the purpose differs.

The comparison yields a simple conclusion. The NucLex contribution, if it is made, would not be a new interaction format. It would be an editorial and governance layer that selects, curates, and versions interaction records in the nuclear medicine domain, expresses them where possible in existing formats, and adds the quantitative, uncertainty-aware, provenance-preserving content that a dosimetry or kinetic model needs and that no single existing resource delivers together.

Nuclear medicine relevance

Consider a radioligand designed to bind a cell-surface target expressed on tumor cells, labeled with a therapeutic beta-emitting radionuclide. The physician wants to know whether a particular patient's tumor will retain the agent long enough to deliver a useful absorbed dose; the radiochemist wants to know whether the labeled construct binds as well as the cold one; the modeler wants to build a kinetic model and then a dosimetry estimate. Each needs the same interaction record and each needs different parts of it.

The entity records tell them what the ligand is and what the target is. The interaction record tells them that the labeled ligand binds the target with an equilibrium dissociation constant measured under stated conditions, that binding is followed by internalization with a stated half-time in a stated cell line, and that a competing endogenous substrate was or was not present in the assay. It tells them the uncertainty of each value and where it came from.

Now ask what a kinetic model would need beyond that record. A compartment model of ligand in blood, interstitial space, tumor cell surface, and tumor cell interior requires association and dissociation rate constants, not just an equilibrium constant; an internalization rate; a target density per cell or per gram of tissue, which depends on the patient's tumor rather than on the cell line; a blood clearance profile; and the physical decay constant of the radionuclide, which is a property of molecule A alone. A dosimetry calculation on top of that model requires the time-integrated activity in each source region and the absorbed dose per unit of time-integrated activity for each source-target pair, in the nomenclature MIRD pamphlet 21 standardized (Bolch et al., 2009). None of these later quantities are interaction records. The interaction record supplies some of the inputs, with the conditions under which they were measured, and leaves visible the assumptions a modeler must make to move from a membrane preparation at 4 degrees Celsius to a living tumor at body temperature. That visibility is the proposal's entire purpose. It does not replace a measurement in the patient and it does not produce a dose.

Disagreements and limitations

The strongest objection to the proposal is that it duplicates PSI-MI, BioPAX, and SBML annotation practice. The monograph has tried to state that objection fairly. The reply is that these standards were built for proteomics, pathway biology, and systems modeling respectively, and that nuclear medicine's particular needs (the labeled construct as a distinct entity, the radionuclide's physical constants as a separate input, dosimetry nomenclature, and clinical review of parameter applicability) are not served by any one of them alone. Whether an extension or a new layer is the right answer is unresolved.

A second objection concerns feasibility. Curating interaction records with full context and uncertainty is slow and expensive; Reactome's model requires named curators and reviewers for every reaction. NucLex has no such corpus and has not demonstrated the capacity to build one. The claim here is a design, not a capability.

A third limitation is the risk of false precision. A record that carries a dissociation constant to three significant figures with a stated standard deviation looks authoritative. If the assay was unrepresentative of the clinical setting, the precision is irrelevant and the applicability note is the only protection. Reviewers should treat the applicability note as the most important field in the record.

Finally, the monograph has not addressed interactions that are not molecular: a radiopharmaceutical interacting with a kidney, a therapy interacting with a prior therapy, a patient interacting with a schedule. The founder's requirement was stated for molecules. Extending the concept further would need its own argument.

References

Ashburner M, Ball CA, Blake JA, et al. Gene Ontology: tool for the unification of biology. Nature Genetics. 2000;25(1):25-29.
Bergmann FT, Adams R, Moodie S, et al. COMBINE archive and OMEX format: one file to share all information to reproduce a modeling project. BMC Bioinformatics. 2014;15:369.
Bolch WE, Eckerman KF, Sgouros G, Thomas SR. MIRD pamphlet No. 21: a generalized schema for radiopharmaceutical dosimetry: standardization of nomenclature. Journal of Nuclear Medicine. 2009;50(3):477-484.
Demir E, Cary MP, Paley S, et al. The BioPAX community standard for pathway data sharing. Nature Biotechnology. 2010;28(9):935-942.
Hermjakob H, Montecchi-Palazzi L, Bader G, et al. The HUPO PSI's Molecular Interaction format: a community standard for the representation of protein interaction data. Nature Biotechnology. 2004;22(2):177-183.
Hucka M, Finney A, Sauro HM, et al. The systems biology markup language (SBML): a medium for representation and exchange of biochemical network models. Bioinformatics. 2003;19(4):524-531.
Jassal B, Matthews L, Viteri G, et al. The reactome pathway knowledgebase. Nucleic Acids Research. 2020;48(D1):D498-D503.
Kerrien S, Orchard S, Montecchi-Palazzi L, et al. Broadening the horizon: level 2.5 of the HUPO-PSI format for molecular interactions. BMC Biology. 2007;5:44.
Le Novère N, Finney A, Hucka M, et al. Minimum information requested in the annotation of biochemical models (MIRIAM). Nature Biotechnology. 2005;23(12):1509-1515.
SBML. What is SBML. https://sbml.org/documents/what-is-sbml/ (accessed 10 October 2026).
Szklarczyk D, Gable AL, Lyon D, et al. STRING v11: protein-protein association networks with increased coverage, supporting functional discovery in genome-wide experimental datasets. Nucleic Acids Research. 2019;47(D1):D607-D613.
W3C. PROV-O: The PROV Ontology. W3C Recommendation, 30 April 2013. https://www.w3.org/TR/prov-o/
NucLex Detailed Discussion Record, sections 8, 9 and 35 (internal project document, revised 9 October 2026).