Skip to content

Psychiatric and Neurodevelopmental Curation SOP: Circuit-Level Mechanisms

Curator-facing SOP for representing psychiatric and neurodevelopmental disorders in dismech using the existing pathophysiology model. Resolves #1448 — the question of whether circuit-level disorders (e.g. ADHD) need a schema extension.

Short answer: no schema extension is needed. Circuit-level mechanisms, intermediate cognitive constructs, and observable behavioural phenotypes are all expressible with the existing pathophysiology / phenotypes model plus the standard ontology bindings (GO, CL, UBERON, HPO). This document captures the recurring modeling decisions so future psychiatric entries converge on the same patterns the ADHD prototype established.

The general curation mechanics (evidence, references, ontology terms) live in CLAUDE.md and .claude/skills/. This document covers only the modeling-decision questions specific to circuit-level disease: how to express a "circuit," where to place an intermediate cognitive construct, and when to keep an axis as a subtype facet rather than splitting into a separate disease entry.

Curation unit: one entry per disorder, not per circuit

A psychiatric dismech entry is a curation unit centered on one disorder's mechanism graph, not one circuit per file. A single entry should be able to chain from genetic liability → neurotransmitter signaling → circuit-level dysregulation → cognitive/behavioural construct → clinical phenotype, all within the same pathophysiology list.

Keep as one entry when the phenotypes and circuit findings share a single diagnostic concept (DSM/ICD/MONDO disease class), even if multiple circuits or neurotransmitter systems are involved. ADHD, schizophrenia, autism, and Tourette syndrome are each one entry that internally chains across several circuits and signaling axes.

Split into separate entries only when the causal program is genuinely distinct — different diagnostic concept, different cell-of-origin or neurotransmitter axis, different therapy program. ADHD vs. Tourette vs. conduct disorder are separate files; ADHD's "Combined / Inattentive / Hyperactive-Impulsive" presentations are subtype facets within the ADHD entry, not separate files.

Representing circuit-level mechanisms

Circuits are represented as ordinary pathophysiology nodes whose circuit semantics come from the existing ontology-bound slots, chained via downstream causal edges. The "circuit" is expressed three ways simultaneously:

  1. Multi-region locations on the node — e.g., a frontostriatal node carries both UBERON:0000451 prefrontal cortex and UBERON:0002435 striatum. The two regions on one node mean "this mechanism spans these anatomical structures."
  2. Cell types via CL that participate in the circuit — e.g., CL:0000617 GABAergic neuron, CL:1001474 medium spiny neuron, CL:0008031 cortical interneuron, CL:0000700 dopaminergic neuron, CL:0008025 noradrenergic neuron.
  3. Biological processes via GO that describe what's happening at the synaptic / signaling layer — e.g., GO:0050804 modulation of chemical synaptic transmission (with modifier: ABNORMAL), GO:0014046 dopamine secretion (modifier: DECREASED), GO:0048243 norepinephrine secretion (modifier: DECREASED).

The circuit's topology (which region projects to which) is then expressed through downstream edges between pathophysiology nodes, not by inventing a "circuit" slot. The ADHD entry's worked chain:

flowchart TD
    A["Catecholaminergic Signaling Deficit in Prefrontal Cortex<br/>(PFC; CL:0000700, CL:0008025)"]
    B["Prefrontal Cortex Circuit Weakness<br/>(PFC; CL:0008031)"]
    C["Frontostriatal Circuit Dysregulation<br/>(PFC + striatum; CL:0000617)"]
    D["Executive Function and Attention Regulation Impairment"]
    E["Short Attention Span / Hyperactivity / Impulsivity<br/>(phenotypes)"]
    A -->|downstream| B
    B -->|downstream| C
    C -->|downstream| D
    D -->|downstream| E

Tourette syndrome demonstrates the same pattern for the cortico-striato-thalamo-cortical (CSTC) loop: one node carries prefrontal cortex + striatum + dorsal-plus-ventral thalamus (UBERON:0001897) as multi-region locations plus medium spiny neurons and GABAergic interneurons, and chains downstream to Dopamine-Modulated Tic Circuit Output and on to the tic phenotypes.

When you introduce a new circuit node, prefer reusing GO terms already in use across the psychiatric entries (GO:0050804 modulation of chemical synaptic transmission and GO:0007268 chemical synaptic transmission are the most common substrates). Add a modifier (ABNORMAL, INCREASED, DECREASED) rather than minting a new ontology term.

Intermediate cognitive constructs: pathophysiology node and phenotype

A recurring question for psychiatric disorders is whether deficits like executive function, working memory, or salience attribution should be HPO phenotypes or pathophysiology nodes referencing GO.

The worked answer is: both, at different layers. The intermediate cognitive construct is a pathophysiology node referencing GO, and it carries downstream edges to the observable behavioural phenotypes referencing HPO. The construct sits between circuit dysfunction and clinical observation, where it belongs causally.

The ADHD entry is the canonical example:

- name: Executive Function and Attention Regulation Impairment   # pathophysiology, not phenotype
  biological_processes:
  - preferred_term: cognition
    term: {id: GO:0050890, label: cognition}
    modifier: ABNORMAL
  - preferred_term: short-term memory
    term: {id: GO:0007614, label: short-term memory}
    modifier: DECREASED
  locations:
  - preferred_term: prefrontal cortex
    term: {id: UBERON:0000451, label: prefrontal cortex}
  downstream:
  - target: Short Attention Span        # HPO phenotype below
  - target: Hyperactivity and Impulsivity

phenotypes:
- name: Short Attention Span             # observable behaviour, not mechanism
  category: Behavioral
  phenotype_term:
    preferred_term: Short attention span
    term: {id: HP:0000736, label: Short attention span}
  diagnostic: true

Rules of thumb:

  • Pathophysiology + GO when the construct is a mechanism — a process that can be increased/decreased/abnormal and that other circuit nodes can causally drive. Cognition, working memory, salience attribution, response inhibition, sensory gating, social cognition, reward processing — all pathophysiology nodes referencing GO.
  • Phenotypes + HPO when the item is an observable clinical sign the patient or clinician reports — inattention, hyperactivity, impulsivity, motor/vocal tics, hallucinations, social communication deficits. category: Behavioral or Cognitive is appropriate on these phenotypes.

The cognitive-construct node should usually carry locations (the substrate brain region) even though no cell-type or process change is unique to it — this preserves the chain so the pathograph renders cleanly from circuit → construct → phenotype.

If a phenotype itself has both mechanistic and observational facets (e.g. psychotic symptoms is both a state of abnormal salience attribution and a clinical observation), split it: keep the mechanism layer as a pathophysiology node (with GO + UBERON), and keep the symptom as an HPO phenotype downstream.

Diagnostic-presentation subtypes vs. comorbidity entries

Psychiatric disorders frequently have DSM/ICD "presentations" (ADHD Combined / Inattentive / Hyperactive-Impulsive; schizophrenia paranoid / disorganized / catatonic) and frequent comorbid disorders (Tourette syndrome ↔ ADHD ↔ OCD).

  • Presentations of one disorder → has_subtypes within the entry. ADHD's three presentations are subtypes; they share the same mechanism graph and differ only in symptom weighting.
  • Distinct comorbid disorders → separate entries, linked via comorbidities (and phenotype-level entries where comorbid features appear as phenotypes of the index disorder). Tourette syndrome lists ADHD, OCD, anxiety and depression as phenotype-level entries and as comorbidities, but each is also its own dismech file.

If you are unsure: prefer one entry with a subtype facet over a new file. Splitting psychiatric entries by presentation will fragment the mechanism graph without adding biology.

Genetic factors: polygenic liability is a first-class node

Psychiatric and neurodevelopmental disorders are typically polygenic; large GWAS, rare CNV, and de novo mutation contributions coexist. Use:

  • A Polygenic Inherited Liability pathophysiology node at the top of the chain that summarises GWAS-supported liability and points downstream to the molecular and circuit nodes. ADHD, Tourette, autism, and schizophrenia all start their chains this way.
  • A genetic_factors block (or genetics) for the gene-level details (DRD2, GRIN2A, SHANK3, NRXN1, CHD8, HDC, SLITRK1, etc.), each with HGNC term, association strength, and PMID evidence.
  • A Brain Gene-Expression Regulation Effects node when bulk transcriptomic / splicing / methylation findings are part of the story, feeding into the circuit nodes.

Reserve the polygenic / liability framing for the disorder-level node; individual genes belong in genetic_factors, not as standalone pathophysiology nodes — unless the gene's product has its own well-defined mechanistic role and direct downstream edges to circuit nodes.

Evidence patterns for circuit-level claims

Circuit-level pathophysiology claims often draw from review articles, neuroimaging meta-analyses, and TMS/MRI studies. Apply the standard evidence SOP from CLAUDE.md, with these notes:

  • evidence_source: HUMAN_CLINICAL for review articles, neuroimaging meta-analyses, and clinical neurobiology reviews (this is the default for most circuit-level evidence).
  • supports: PARTIAL (with evidence_source: HUMAN_CLINICAL) when an imaging meta-analysis supports circuit dysconnectivity broadly but the abstract does not localize the exact tract the node names — this is common for white-matter findings; do not overclaim. Example:
evidence:
- reference: PMID:XXXXXXXX
  supports: PARTIAL
  evidence_source: HUMAN_CLINICAL
  snippet: "..."
  explanation: "Meta-analysis supports broad frontostriatal dysconnectivity but does not localize to the specific tract named in this node."

Note: supports: (SUPPORT / PARTIAL / REFUTE / NO_EVIDENCE / WRONG_STATEMENT) and evidence_source: (HUMAN_CLINICAL / MODEL_ORGANISM / IN_VITRO / COMPUTATIONAL / OTHER) are independent fields — PARTIAL is only valid on supports:. - Pharmacological-efficacy evidence (e.g. "dopamine-blocking agents are effective for tics") is indirect mechanistic evidence — mark it PARTIAL or SUPPORT with evidence_source: HUMAN_CLINICAL and explain the inference in the explanation field. Don't treat treatment-response as direct circuit evidence. - Imaging-only evidence is usually HUMAN_CLINICAL; iPSC-derived neurons and patient-derived organoid studies are IN_VITRO; knockout mouse circuit findings are MODEL_ORGANISM. Keep these on separate evidence items so each one carries a single evidence_source.

Worked exemplars

When in doubt about a new psychiatric entry, mirror one of these:

Disorder Entry What it demonstrates
ADHD kb/disorders/Attention_Deficit-Hyperactivity_Disorder.yaml Polygenic liability → catecholamine signaling → PFC weakness → frontostriatal → executive function → behavioural phenotypes; DSM presentations as subtypes
Schizophrenia kb/disorders/Schizophrenia.yaml Multiple parallel pathophysiology axes (dopamine / glutamate / GABA / complement-mediated pruning / oligodendrocyte / mitochondrial / BBB) feeding a shared symptom set
Autism spectrum disorder kb/disorders/Autism_Spectrum_Disorder.yaml E/I imbalance as integrating mechanism; neuroinflammation, gut-brain axis, synaptic scaffolding, chromatin remodeling as parallel contributors
Tourette syndrome kb/disorders/Tourette_Syndrome.yaml CSTC circuit modeled as multi-region pathophysiology node; dopamine-modulated tic output as proximal mechanism; tic phenotypes downstream
Conduct / oppositional defiant disorder kb/disorders/Conduct_Disorder.yaml, kb/disorders/Oppositional_Defiant_Disorder.yaml Behavioural-disorder framing alongside neurodevelopmental disorders
  • CLAUDE.md — repo-wide curation conventions (evidence SOP, ontology binding rules, MAXO/CHEBI treatment patterns)
  • docs/cancer-curation-sop.md — analogous SOP for cancer entries (MONDO/NCIT anchoring, subtype axes)
  • docs/pathographs.md — how pathophysiology.downstream edges and phenotypes.sequelae edges render as interactive graphs
  • docs/frequency-evidence-guidelines.md — frequency-qualifier evidence patterns relevant to behavioural and cognitive phenotype frequencies
  • #1448 — this document
  • #1307 — sibling discussion on faceted subtypes vs. separate disease pages