Validation Report: Zhu/Dann T Cell Perturb-seq vs DisMech Autoimmune Entries
Date: 2026-02-14 Pipeline: Zhu, Dann, Ota, Pritchard, Marson et al. (bioRxiv Dec 2025) Method: Genome-scale CRISPRi Perturb-seq in primary CD4+ T cells; 111 regulator clusters tested for enrichment of GWAS disease genes (Open Targets, genetic_evidence >= 0.1)
Summary
| Metric | Value |
|---|---|
| Diseases evaluated | 12 (of 14 tested; autoimmune umbrella excluded, atopic eczema not in KB) |
| Significant cluster-disease pairs (FDR < 0.05) | 146 |
| Unique GWAS genes tested | 408 |
| Genes confirmed in DisMech | 11 (2.7%) |
| Genes novel to DisMech | 405 (99.3%) |
Per-Disease Confirmation Rates
| Disease | DisMech Genes | Pairs | Confirmed | Novel | Rate |
|---|---|---|---|---|---|
| Type 1 diabetes | 10 | 5 | 3 | 34 | 8.1% |
| Hashimoto's thyroiditis | 4 | 7 | 2 | 37 | 5.1% |
| Multiple sclerosis | 6 | 14 | 3 | 65 | 4.4% |
| SLE | 11 | 11 | 2 | 65 | 3.0% |
| Asthma | 6 | 13 | 3 | 110 | 2.7% |
| Rheumatoid arthritis | 9 | 15 | 3 | 139 | 2.1% |
| Psoriasis | 5 | 9 | 1 | 59 | 1.7% |
| Ulcerative colitis | 4 | 16 | 1 | 96 | 1.0% |
| IBD (merged) | 10 | 17 | 1 | 156 | 0.6% |
| Crohn's disease | 6 | 17 | 0 | 98 | 0.0% |
| Ankylosing spondylitis | 3 | 15 | 0 | 74 | 0.0% |
| Celiac disease | 4 | 7 | 0 | 27 | 0.0% |
Confirmed Genes
These DisMech-curated genes were rediscovered by the T cell Perturb-seq pipeline:
| Gene | Confirmed In | Role |
|---|---|---|
| PTPN22 | RA, Hashimoto's, T1D, SLE (12 pairs) | Shared autoimmune susceptibility; T cell signaling |
| CTLA4 | RA, Hashimoto's, T1D (5 pairs) | T cell co-inhibition; immune checkpoint |
| IL2RA | MS, T1D (5 pairs) | IL-2 receptor alpha; Treg homeostasis |
| IL4 | Asthma (4 pairs) | Th2 cytokine; allergic inflammation |
| STAT4 | RA, SLE (2 pairs) | Th1/Th17 signaling |
| IL13 | Asthma (2 pairs) | Th2 cytokine; allergic inflammation |
| IL23R | IBD (2 pairs) | Th17 differentiation |
| IL7R | MS (2 pairs) | T cell homeostasis |
| ORMDL3 | Asthma (1 pair) | ER stress, sphingolipid metabolism |
| TNFAIP3 | Psoriasis (1 pair) | NF-kB negative regulation (A20) |
| TNFRSF1A | MS (1 pair) | TNF receptor; neuroinflammation |
Interpretation
Why is the confirmation rate so low?
The 2.7% rate does not indicate contradiction. It reflects a gene coverage gap in DisMech:
-
DisMech has 3-11 genes per disease (median ~6), curated from classic GWAS papers and textbook genetics. These are the "greatest hits" -- HLA loci, PTPN22, IL23R, NOD2, etc.
-
The pipeline tests hundreds of genes per disease from Open Targets (genetic_evidence_score >= 0.1), including many more recent GWAS discoveries, rare variant associations, and genes with modest effect sizes.
-
The "novel" genes are mostly real. EGR2, BACH2, IRF4, STAT3, TNFAIP3, IKZF1, CD28, GATA3, SMAD3 are well-established autoimmune risk genes in the literature. They are "novel" only in the sense that DisMech hasn't curated them yet.
What does DisMech confirm well?
The confirmed genes cluster into functional categories: - T cell signaling: PTPN22, CTLA4, IL2RA, IL7R, CD28 (partial) - Cytokines: IL4, IL13, IL23R - Transcription factors: STAT4
These are exactly the genes where DisMech has the strongest mechanistic documentation.
What does DisMech miss?
The top "novel" genes (appearing in 7+ diseases) reveal systematic gaps:
| Gene | #Diseases | Function | Gap Type |
|---|---|---|---|
| EGR2 | 12 | T cell anergy/tolerance | Known risk gene, not curated |
| BACH2 | 12 | Treg/effector balance, class switch | Known risk gene, curated only in T1D |
| ETS1 | 11 | T cell development, Th17 | Known risk gene, not curated |
| IRF4 | 10 | Plasma cell differentiation, Th17 | Known risk gene, curated only in Vitiligo |
| STAT3 | 8 | Th17 signaling, acute phase | Known risk gene, not curated |
| IKZF1 | 9 | Lymphocyte development (Ikaros) | Known risk gene, not curated |
| CD28 | 9 | T cell co-stimulation | Known risk gene, not curated |
| GATA3 | 7 | Th2 master regulator | Known risk gene, not curated |
| SMAD3 | 9 | TGF-beta signaling, Treg | Known risk gene, not curated |
| IL10 | 8 | Anti-inflammatory cytokine | Known risk gene, not curated |
Actionable Findings
1. High-Priority Gene Additions
These genes should be added to the genetic sections of multiple dismech entries.
Ranked by number of diseases where they appear as GWAS hits downstream of T cell regulators:
Tier 1 (10+ diseases): EGR2, BACH2, IL2RA, IL7R, TNFAIP3, ETS1, ADO, IRF4 Tier 2 (7-9 diseases): IKZF1, STAT3, CD28, SMAD3, SATB1, NFKBIA, GATA3, CD6, IRF8, IL10, RASGRP1, FOXO1, PTPN22 Tier 3 (5-6 diseases): SOCS1, MAP3K8, RIPK2, BATF, IL2, FAS, BCL2L11, SH2B3
(*) Already curated in some diseases but missing from others.
2. Missing Disease: Atopic Dermatitis/Eczema
Atopic eczema was tested in the pipeline (16 significant pairs) but we have no KB entry. This is a high-priority curation target given: - Cluster 79 (Th17; IRF4, BATF, STAT3): OR=38.1 - Cluster 110: OR=24.3 - Strong overlap with asthma genetics
3. Cluster 79 (Th17 Differentiation) is the Standout
This 6-gene cluster (IRF4, BATF, STAT3, JUNB, IPMK, NUP188) shows the strongest enrichment across diseases: - Crohn's: OR=58.2 - Atopic eczema: OR=38.1 - Psoriasis: OR=26.9 - IBD: OR=24.0 - MS: OR=16.9 - Autoimmune (general): OR=9.0
DisMech should ensure Th17 differentiation is prominently featured in the pathophysiology of Crohn's, Psoriasis, and Ankylosing Spondylitis.
4. Schema Gap: Gene-Disease Specificity
Many genes (PTPN22, IL2RA, BACH2) are shared across autoimmune diseases but DisMech only records them in 1-4 entries. The pipeline data provides evidence for which diseases each gene is relevant to, with enrichment statistics. Consider: - Systematic addition of pleiotropic autoimmune genes across all relevant entries - Recording the T cell regulatory context (which cluster, which condition)
Methods
Data Sources
- Zhu/Dann enrichment results:
cluster_autoimmune_enrichment_results.suppl_table.csvfrom GitHub - Disease gene sets: Open Targets Platform, genetic_evidence_score >= 0.1
- dismech entries: 12 autoimmune disorder YAML files from
kb/disorders/
Matching Logic
- Gene symbols from enrichment
intersecting_genesmatched against dismechgenetic[].nameandpathophysiology[].genes[].name - Exact string match (case-sensitive)
- IBD umbrella mapped to union of Crohn's Disease + Ulcerative Colitis genes
Limitations
- Only matches gene symbols in structured fields; does not search free-text descriptions
- DisMech uses some non-standard gene names (e.g., "HLA-B27" vs "HLA-B", "BCR-ABL1")
- Does not assess whether DisMech's pathophysiology narratives describe the gene's mechanism
- The pipeline's gene sets come from Open Targets (which includes GWAS + rare variant evidence), not just the T cell Perturb-seq results directly
Data Files
- Validation script:
scripts/validate_tcell_clusters.py - Enrichment data:
docs/assets/zhu_dann_2025/cluster_autoimmune_enrichment_results.csv - Cluster annotations:
docs/assets/zhu_dann_2025/clustering_results_and_annotations.csv