End-to-end differential expression from a single question
A walk-through of how the unified MCP server runs a complete differential expression analysis of diabetic nephropathy in ARCHS4 from one chat — including running both pooled and study-matched meta-analysis modes so the user can see what each method reveals.
An end-to-end DE workflow — disease ontology resolution, ARCHS4 sample classification, statistical testing, enrichment — executed from one chat. Comparing pooled and study-matched modes revealed an interferon signaling signal (OAS2, RSAD2) that the simpler pooled analysis would have missed.
The scenario
Run a pooled differential expression analysis of diabetic nephropathy using ARCHS4. Then try the study-matched meta-analysis. Tell me your steps as you go and finish with a report on the results.
How the assistant approached it
The MCP server orchestrated four tools in sequence:
- Disease ontology resolution.
resolve_disease_ontologymapped “diabetic nephropathy” to MONDO:0005016 (diabetic kidney disease) with exact confidence, no subtype expansion. - Sample discovery.
find_samplesqueried ARCHS4 metadata with ontology-enhanced search and applied LLM-generated regex patterns (diabetic nephropathy, DKD, DN, diabetic glomerulopathy, STZ kidney, db/db kidney) filtered to kidney/renal/glomerular/tubular/podocyte tissue. Per-study LLM classification labeled each sample as test or control. - Differential expression.
differential_expressionran first in pooled mode (Mann-Whitney U with Benjamini-Hochberg FDR), then in study-matched mode (per-study DE combined via Stouffer's weighted Z), with FDR < 0.01 and |log2FC| > 2.0. - Enrichment.
enrichment_analysisran g:Profiler over GO, KEGG, and Reactome on the pooled DE gene set.
Findings
Sample discovery
- 174 DN test samples across 13 ARCHS4 studies
- 555 kidney control samples across 161 studies
- 9 studies contributed both test and control samples
Top contributing studies (test / control samples):
| Study | Test | Control |
|---|---|---|
| GSE175759 | 62 | 19 |
| GSE142025 | 35 | 0 |
| GSE162830 | 22 | 8 |
| GSE185011 | 20 | 5 |
| GSE204880 | 6 | 5 |
| GSE199437 | 6 | 3 |
The system flagged study-matched mode as the recommended methodology but executed pooled first as requested, then ran study-matched for comparison.
Pooled differential expression
100 test samples (10 studies) vs. 53 control samples (8 studies), 19,282 genes tested. Five genes significant, all downregulated:
| Gene | log2FC | Adj. p |
|---|---|---|
| DUSP1 | −2.78 | 2.5e-10 |
| FOS | −3.39 | 1.9e-7 |
| FOSB | −2.66 | 1.8e-7 |
| NR4A1 | −2.61 | 5.0e-7 |
| EGR1 | −2.35 | 2.4e-6 |
All five are immediate-early response transcription factors. The coordinated downregulation suggests suppression of immediate-early transcriptional responses in DN kidney — loss of adaptive stress signaling and MAPK/AP-1 deregulation.
Enrichment (pooled DE genes)
- Reactome: NGF-stimulated transcription (p = 7.5e-5); nuclear events: kinase and TF activation (p = 3.0e-4); NTRK1/TRKA signaling (p = 2.0e-3)
- GO Molecular Function: DNA-binding transcription activator activity, RNA Pol II-specific (p = 3.1e-4)
- GO Biological Process: cellular response to chemical stimulus (p = 0.010); response to hormone (p = 0.013)
- KEGG: MAPK signaling pathway (p = 0.021)
Study-matched meta-analysis
Only three studies had both test and control samples meeting the minimum threshold (GSE175759, GSE199437, GSE204880; 71 test / 27 control; 15,943 genes tested in the intersection). Seven genes significant — five down, two up:
| Gene | log2FC | Adj. p | Direction |
|---|---|---|---|
| FOSB | −2.73 | 4.7e-8 | down |
| EGR1 | −2.47 | 4.5e-8 | down |
| FOS | −3.24 | 9.1e-6 | down |
| DUSP1 | −2.16 | 5.0e-6 | down |
| NR4A1 | −2.33 | 8.0e-6 | down |
| OAS2 | +2.29 | 0.002 | up |
| RSAD2 | +2.09 | 0.005 | up |
The two upregulated hits are interferon-stimulated genes: OAS2 (2′-5′-oligoadenylate synthetase 2) and RSAD2 (Viperin), both canonical antiviral / type I interferon response genes — a signal completely missed by the pooled analysis.
Pooled vs. study-matched
| Feature | Pooled | Study-matched |
|---|---|---|
| Test samples | 100 (10 studies) | 71 (3 studies) |
| Control samples | 53 (8 studies) | 27 (3 studies) |
| Genes tested | 19,282 | 15,943 |
| Significant genes | 5 | 7 |
| All downregulated? | yes (5/5) | no — 2 upregulated |
Both methods recover the same five immediate-early genes with consistent effect sizes. Study-matched additionally surfaces OAS2 and RSAD2 by controlling for batch effects within each study before combining via Stouffer's weighted Z — a signal obscured when samples are pooled across 10 platforms.
Biological interpretation
The combined picture suggests two coordinated processes in DN kidney:
- Suppression of the immediate-early transcriptional response (FOS, FOSB, EGR1, NR4A1, DUSP1) — loss of adaptive stress signaling and MAPK pathway dampening.
- Activation of innate immune / interferon signaling (OAS2, RSAD2) — consistent with established inflammatory mechanisms in DN progression.
Caveats
- Stringent thresholds (FDR < 0.01, |log2FC| > 2.0) yield small gene lists.
- Pooled mode mixes samples across 10 platforms, introducing potential batch effects.
- Only three studies met the matched-controls minimum for study-matched mode; GSE175759 dominates (62/71 test, 21/27 control).
- Immediate-early genes (FOS, EGR1) are sensitive to tissue processing delays — care is needed in interpreting their downregulation as DN-specific.
- Enrichment analysis returned empty for the study-matched gene set (too few genes split across directions).
Bottom line
The MCP server resolved a disease term, classified hundreds of ARCHS4 samples, ran two complementary DE methods, and produced a coherent biological interpretation — in one chat. The pooled-vs-matched comparison illustrates the broader value: the right method choice is rarely obvious in advance, and an orchestrated pipeline that runs both lets the user see the methodology trade-off instead of guessing at it.