Skip to content

Plan: DisMech as Validation Resource for Causal Gene-to-Trait Analyses

The Idea

Given outputs from causal modeling pipelines (GWAS + Perturb-seq → gene programs → traits), can DisMech's curated pathophysiology knowledge explain, validate, or contextualize the discovered relationships?

This positions DisMech as a benchmark/interpretation layer rather than a data store.

Background: The Ota et al. Methodology

The paper "Causal modelling of gene effects from regulators to programs to traits" (Ota et al., Nature 2025) builds causal graphs:

Gene (regulator) --[β]--> Program --[effect]--> Trait/Phenotype
                          ↑
Gene (member) ----[γ]----/

Key data types: - γ (gamma): Gene effect on trait from LoF burden tests (UK Biobank, 454K participants) - β (beta): Regulatory effect from Perturb-seq knockdowns (K562 cells, 9,498 genes) - Programs: Co-expressed gene modules (60 identified via consensus NMF)

Key outputs: - Program-trait effect sizes with directions - Gene-program regulatory relationships - Causal chains explaining how genes affect traits via biological programs - 73% accuracy in predicting effect directions for top GWAS hits

Use Cases for DisMech

1. Validate Discovered Gene-Program-Trait Relationships

Question: Does a computationally discovered pathway match curated knowledge?

Pipeline Output DisMech Query
GATA1 → erythroid program → MCH Do blood disorders document GATA1 affecting hemoglobin via erythropoiesis?
BCL2 → apoptosis program → lymphocyte count Do immune disorders link BCL2 to apoptosis affecting lymphocytes?

Validation levels: - CONFIRMED: Gene, process, and phenotype all documented with evidence - PARTIAL: Some elements present but incomplete chain - NOVEL: Not in DisMech (curation candidate) - CONTRADICTED: DisMech documents opposite effect

2. Explain Programs via Pathophysiology

Question: What does "Program 17" actually mean in disease context?

Pipeline outputs are anonymous gene modules (Program 1, Program 2, ...). DisMech can provide clinical/mechanistic context:

Program GO Annotation DisMech Diseases Clinical Context
GO:0007049 (cell cycle) CML, Retinoblastoma "Uncontrolled proliferation drives tumor growth"
GO:0006915 (apoptosis) SLE, Autoimmune diseases "Defective clearance of apoptotic cells triggers autoimmunity"
GO:0030218 (erythrocyte differentiation) Sickle Cell, Thalassemia "Ineffective erythropoiesis causes anemia"

3. Prioritize Novel Findings

Question: Which discovered relationships are genuinely novel vs. already known?

Given 500 gene-trait associations from a pipeline, DisMech can partition into: - Known: Already documented (low priority for follow-up) - Novel: Not in literature (high priority for experimental validation) - Contradicted: Conflicts with literature (needs investigation)

4. Identify Mechanistic Gaps in DisMech

Question: Where does computational analysis find things missing from curated knowledge?

If pipeline discovers "Gene X → autophagy → Trait Y" but DisMech's Disease Y entry lacks autophagy in pathophysiology, this flags a curation gap.

Feedback loop: Pipeline discoveries improve DisMech coverage.

5. Cross-Disease Program Analysis

Question: Do different diseases share causal programs?

Shared Program Diseases Implication
Inflammation (GO:0006954) RA, Crohn's, Psoriasis Common therapeutic targets
Fibrosis (GO:0030198) IPF, Liver Cirrhosis, Systemic Sclerosis Shared mechanism despite different organs

Data Requirements

From Ota et al. Pipeline

  1. Program definitions: Gene lists per program + GO annotations
  2. Program-trait effects: Effect sizes and directions for each program-trait pair
  3. Gene-program relationships: Which genes regulate which programs (β values)
  4. Gene-trait effects: Direct LoF burden test results (γ values)

Note: Check if supplementary tables from paper are publicly available.

From DisMech

  1. Gene mentions: Genes in genetic and pathophysiology.genes sections
  2. Process annotations: GO terms in pathophysiology.biological_processes
  3. Phenotype annotations: HP terms in phenotypes
  4. Cell type context: CL terms in pathophysiology.cell_types
  5. Evidence: PMIDs supporting each relationship

Mapping Requirements

Pipeline Entity DisMech Entity Mapping Strategy
Gene symbol GeneDescriptor Direct match or HGNC lookup
Program GO term BiologicalProcessTerm Exact or ancestor match via GO hierarchy
Trait Phenotype HP term Manual curation of trait-to-HPO mappings
Cell type CellTypeTerm Match K562 → CL:0000255 (erythroid)

Evaluation Criteria

Quantitative Metrics

  1. Coverage: % of pipeline gene-trait pairs that DisMech can evaluate
  2. Confirmation rate: % of top pipeline hits confirmed by DisMech
  3. Novelty rate: % of pipeline discoveries not in DisMech
  4. Contradiction rate: % of pipeline results conflicting with DisMech

Qualitative Assessment

  1. Do confirmed findings have strong evidence (experimental PMIDs)?
  2. Are novel findings biologically plausible given DisMech context?
  3. Can contradictions be resolved by examining evidence quality?

DisMech Gaps to Address

Current Strengths

  • 55+ disorders with curated pathophysiology
  • GO/HP/CL term bindings enabling semantic matching
  • Evidence with PMIDs for traceability
  • Structured pathophysiology with biological processes and cell types

Gaps

Gap Impact Remediation
Inconsistent modifier usage Cannot validate effect directions Curation pass to add INCREASED/DECREASED
Genes often missing from pathophysiology Low gene coverage Add genes to pathophysiology entries
No explicit gene→process links Cannot trace causal chains Add gene slot to BiologicalProcessDescriptor
Blood disorders underrepresented Poor coverage for Ota et al. test case Prioritize Thalassemia, Polycythemia, etc.

Pilot: Blood Traits

The Ota et al. paper focuses on three blood traits: - MCH (mean corpuscular hemoglobin) - RDW (red cell distribution width) - IRF (immature reticulocyte fraction)

Relevant DisMech Disorders

Disorder Relevance Current Coverage
Sickle Cell Disease Hemoglobin, erythrocytes Good - has GO terms, cell types, evidence
Autoimmune Hemolytic Anemia Hemolysis, RBC destruction Moderate
Fanconi Anemia Bone marrow failure Moderate
G6PD Deficiency RBC fragility Check coverage
Hemochromatosis Iron metabolism Check coverage

Missing Disorders to Add

  • Thalassemia (alpha and beta)
  • Polycythemia vera
  • Iron deficiency anemia
  • Aplastic anemia
  • Myelodysplastic syndromes

Output Formats

1. Validation Report (per gene-program-trait)

gene: GATA1
program: erythroid_differentiation
program_go: GO:0030218
trait: mean_corpuscular_hemoglobin
effect_direction: positive

validation:
  status: CONFIRMED
  matching_disorders:
    - disease: Sickle Cell Disease
      pathophysiology_entry: "Red Blood Cell Sickling"
      process_match: GO:0030218 (erythrocyte differentiation)
      gene_mentioned: false  # gap identified
      evidence: PMID:24277079
  confidence: MEDIUM
  notes: "Process confirmed, gene not explicitly mentioned"
  curation_action: "Add GATA1 to SCD pathophysiology"

2. Summary Statistics

Total pipeline relationships evaluated: 180
  - CONFIRMED: 73 (41%)
  - PARTIAL: 45 (25%)
  - NOVEL: 52 (29%)
  - CONTRADICTED: 10 (6%)

Coverage by program type:
  - Cell cycle programs: 85% evaluable
  - Erythroid programs: 92% evaluable
  - Autophagy programs: 34% evaluable (gap)

3. Curation Candidates

Ranked list of novel findings with high effect sizes that warrant adding to DisMech.

4. Gap Report

Which GO terms from programs lack DisMech coverage, prioritized by frequency in pipeline results.

Implementation Phases

Phase 1: Data Assembly

  • Obtain Ota et al. supplementary data
  • Export DisMech blood disorders to queryable format
  • Create trait-to-HPO mapping for MCH, RDW, IRF

Phase 2: Proof of Concept

  • Manual validation of 20 top gene-program-trait relationships
  • Document matching logic and edge cases
  • Assess DisMech coverage and gaps

Phase 3: Systematic Evaluation

  • Automate matching across all pipeline outputs
  • Generate validation report
  • Quantify confirmation/novelty/contradiction rates

Phase 4: Bidirectional Improvement

  • Use novel findings to improve DisMech coverage
  • Re-run validation to measure improvement
  • Publish methodology and results

Open Questions

  1. Trait mapping: How to map blood lab values (MCH) to HPO terms? Direct terms exist (HP:0025066 "Decreased mean corpuscular hemoglobin") but may need expert review.

  2. Cell type specificity: Ota et al. used K562 (erythroleukemia line). How generalizable are programs to other cell types in DisMech pathophysiology?

  3. Effect direction validation: DisMech has modifier but it's underused. How much curation effort to systematically add directions?

  4. Semantic matching depth: Should GO term matching use exact match, or traverse hierarchy (e.g., "erythrocyte differentiation" matches "hematopoiesis")?

  5. Evidence quality weighting: Should we weight DisMech evidence by experimental type (IDA > TAS > NAS)?

References

Assets

  • Paper PDF: docs/assets/Causal modelling of gene effects from regulators to programs to traits.pdf
  • Table S1: docs/assets/Ota_2025_TableS1.xlsx - Program annotations (60 programs)
  • Curated annotations (e.g., "Hemoglobin synthesis", "Cell cycle (S phase)", "Autophagosome")
  • Representative GO terms
  • Top 10 genes per program
  • Representative transcription factors
  • Table S2: docs/assets/Ota_2025_TableS2.xlsx - Trait definitions (54 traits)
  • RBC traits: MCH, MCV, RDW, reticulocyte count, hemoglobin, etc.
  • Other blood traits: platelet count, WBC, lymphocyte count, etc.
  • Serum biomarkers: albumin, cholesterol, glucose, bilirubin, etc.
  • Anthropometric: BMI, height, bone mineral density

Key Programs for Blood Trait Validation

Program Annotation GO Term Relevance to DisMech
P4 Cell cycle (S phase, DNA replication) DNA replication CML, Retinoblastoma
P6 Cell cycle (G2M checkpoint) G2M checkpoint Cancers
P16 Autophagosome autophagosome Neurodegeneration, inflammation
P27 Platelet activation (Mega, IKZF1) platelet activation Immune thrombocytopenia
P28 RBC (glycoproteins) HEME_METABOLISM Sickle Cell, Thalassemia
P35 TNF signaling (stress response) TNFA_SIGNALING_VIA_NFKB Autoimmune diseases
P40 Hemoglobin synthesis HEME_METABOLISM Sickle Cell, Thalassemia, Anemia
P45 TNF signaling (apoptosis) TNFA_SIGNALING_VIA_NFKB SLE, RA

Data Not Yet Available

The bioRxiv preprint only includes Tables S1 and S2. Additional data mentioned in the paper: - Table S3: Program-trait effect sizes (needed for quantitative validation) - Gene-level burden test results (γ values) - GitHub repository with analysis code (not yet public)

These may become available with the Nature publication or upon request to authors.