Why Science? | Molecular Interactions Controlled Vocabulary and Comparable Interaction Evidence

Molecular-interaction data can look like a simple line between two nodes, yet the scientific meaning depends on who the participants are, how the interaction was detected, which roles they had, what biological system was used and whether the record represents direct binding, a complex, colocalisation or another kind of evidence. The HUPO Proteomics Standards Initiative maintains the Molecular Interactions Controlled Vocabulary, commonly called PSI-MI. Its official repository supplies OBO and OWL products, a public issue process for new entities and instructions for reviewed contributions; it states that the EBI Ontology Lookup Service reads the master vocabulary weekly. PSI-MI terms make interaction records easier to exchange, but a shared term does not make assays or biological conclusions equivalent. This article supports Primary Science, PSLE Science, Secondary Science, O-Level Science, O-Level Biology, data literacy and STEM education. It is not medical advice, a laboratory protocol, a database endorsement, an admission promise or a career guarantee.

Related eduKate reading: Relation Ontology and shared scientific relationships; Evidence and Conclusion Ontology and computable evidence; Cell Ontology and interoperable cell types; Ontology for Biomedical Investigations and reproducible experiments; Science Learning Hub; Education Hub; STEM Education; Careers by Subject Capability.

Primary and official sources checked on 13 October 2026: Official HUPO-PSI Molecular Interactions Controlled Vocabulary repository, checked 13 October 2026; Official PSI-MI OBO vocabulary file; Official PSI-MI OWL vocabulary file; HUPO Proteomics Standards Initiative; 2026 Singapore–Cambridge O-Level Biology syllabus; MOE G2/G3 Lower Secondary Science teaching and learning syllabus; 2026 MOE G2 Computing syllabus. Ontologies, definitions and release products can change, so preserve dated documentation, exact source records and file versions.

How an experimental observation becomes comparable molecular-interaction evidence

An interaction record is more than two names. Protein isoforms, complexes, nucleic acids, small molecules and experimental constructs can occupy participant positions, and their identities determine what the edge means.

The official PSI-MI repository provides stable vocabulary terms in OBO and OWL forms. Identifiers let databases exchange interaction meaning without depending only on interface labels.

Detection method is part of evidence. Affinity capture, two-hybrid approaches, cross-linking, imaging and biophysical assays observe different aspects of molecular association and carry different artefacts.

Experimental role and biological role are not interchangeable. A bait or prey label describes the assay configuration, while enzyme, substrate or other biological roles concern the interpreted system.

Interaction type needs careful reading. Physical association, direct interaction and membership in a larger complex support different strengths of mechanistic statement.

Participant identification methods matter because an instrument or antibody can misidentify, group or miss molecular forms. Keep identification evidence beside interaction detection evidence.

Stoichiometry and participant count are not always determined by a positive assay. A pairwise screen can show compatibility without reconstructing the native complex.

Negative results need assay sensitivity and controls. Failure to detect an interaction in one system does not establish that it never occurs in another tissue, compartment or condition.

The repository accepts reviewed term requests through issues and publishes the vocabulary from a maintained master branch. Release or commit identity therefore belongs in reproducible data exchange.

Network diagrams can merge many records into one edge. The graph should preserve the number, type, source and confidence of supporting experiments rather than presenting one uniform truth.

Co-occurrence, co-expression and colocalisation can motivate interaction hypotheses but should not be rewritten as direct molecular binding without suitable evidence.

The optimistic value of PSI-MI is a testable network: people and software can compare interaction records while retaining the experimental vocabulary needed to challenge each edge.

Start with stable participant identities and molecular forms. A gene symbol alone may refer to several proteins, isoforms, cleavage products or engineered constructs.

Record organism and biological material. An interaction in a purified system, cultured cell or tissue can support different claims about native biology.

Describe constructs, tags and mutations. Experimental additions can enable detection while changing localisation, folding or binding behaviour.

Preserve assay direction. Bait and prey configurations, capture order and reporter architecture can reveal or hide interactions asymmetrically.

Use positive, negative and technical controls to show that the signal depends on the intended participants and method.

Record whether an edge is directly observed, curator-inferred, imported or computationally predicted. Predictions are hypotheses until experimentally supported.

Map local method labels to PSI-MI with explicit scope. A laboratory phrase may be broader or narrower than the controlled term selected by a database.

When combining databases, deduplicate source experiments rather than counting repeated imports as independent replication.

For classroom learning, students can compare three kinds of evidence for the same proposed edge and decide which claim each method can actually support.

A reader-ready interaction note names participants, forms, roles, interaction type, method, system, controls, source and limitation in plain language.

Binary edges can be generated from larger complexes by several expansion rules. Spoke and matrix models create different graphs, so the transformation method must be recorded.

A direct assay can establish physical contact under test conditions without proving that the event occurs in the relevant organism, cell type or physiological state.

Affinity purification can recover stable and indirect complex members. Additional evidence is needed before assigning direct pairwise binding among all captured proteins.

Reporter assays can depend on protein localisation, folding and expression in the host system. A negative result may reflect assay incompatibility rather than biological absence.

Mass-spectrometry identification includes peptide assignment, protein inference and false-discovery control. The interaction edge should remain linked to these analytical decisions.

Antibody-based methods need reagent identity and validation because cross-reactivity can attach the wrong participant to an otherwise convincing signal.

Networks are shaped by research attention. Highly studied proteins often have more recorded edges, so degree does not automatically measure biological importance.

Context-specific interactions can depend on modification, ligand, compartment, cell state or developmental stage. A context-free edge may be useful for search but too broad for mechanism.

Machine-learning evaluation must prevent shared source papers, complexes or near-duplicate interactions from leaking across training and test sets.

Vocabulary migration should identify obsolete or replaced terms and rerun method, role and interaction-type validation. Labels alone are not a safe migration key.

Conflicting records deserve representation. Different assays, conditions or curatorial interpretations can produce disagreement that should remain visible rather than averaged away.

The strongest PSI-MI workflow is reversible: every edge can be traced through vocabulary terms to participants, experimental record, source publication, curator and release.

A high-quality interaction exchange should preserve an evidence bundle for every edge. Recommended fields include participant identifiers and molecular forms, organism, expression system, cellular compartment, construct details, experimental and biological roles, interaction type, detection method, participant-identification method, controls, confidence method, source experiment, publication, curator, database accession, vocabulary release and any complex-expansion rule. Predictions and inferred edges should occupy separate assertion fields. If several databases import the same paper, the bundle makes deduplication possible without erasing independent curation. If two experiments disagree, the system can show the conflicting contexts instead of choosing a single winner. This is the difference between an attractive network and a scientifically reviewable one.

For a validation drill, take one proposed protein pair and represent evidence from a capture assay, a proximity image and a computational prediction. Assign the narrowest PSI-MI terms that match each record, then write the strongest sentence each record supports. Next, merge them into a network edge while keeping the evidence list inspectable. Ask whether any experiment established direct contact, native context, stoichiometry or functional consequence. Add a negative result from a different cell type and resist treating it as a universal contradiction. The exercise shows how controlled vocabulary improves comparison without flattening methods. A reliable graph should let a reader open the line, see the underlying records and understand why the final interpretation remains bounded.

Before publication, an interaction database can sample edges from several methods and ask whether the graph preserves participant form, assay, roles, controls, source and assertion type. Any edge that cannot be traced should be quarantined from mechanistic summaries until its evidence bundle is repaired. Broad network views can still include it with an explicit status. This layered approach supports discovery without presenting uncertain records as confirmed direct binding. It also gives software tests concrete expectations: direct-interaction filters, complex-membership filters and prediction filters should return different, explainable sets from the same underlying collection. That difference matters.

A four-stage evidence ladder for PSI-MI-backed interaction records

Stage	Question	Evidence	Responsible output
Identify	Which physical or molecular entities participated?	Stable participant identifiers, forms, organism and cellular context	Bounded participant records
Describe	What interaction and method terms apply?	PSI-MI identifiers, definitions, roles and vocabulary release	A machine-readable interaction record
Support	What experiment justifies the edge?	Detection method, construct, controls, source and curator provenance	Traceable evidence
Interpret	Which biological conclusion is warranted?	Replication, orthogonal methods, context and network validation	A calibrated interaction claim