Generates complete dual-disease transcriptomic + machine learning research designs from a user-provided disease pair. Use when users want to identify shared DEGs, common hub genes, cross-disease biomarkers, or shared molecular mechanisms between two diseases using public GEO data. Triggers: "shared biomarker study for two diseases", "dual-disease transcriptomic ML paper", "identify common DEGs between disease A and B", "cross-disease hub gene discovery", "shared DEG + PPI + ROC design", "immune infiltration shared biomarker", or "I want to study disease X and Y together". Always outputs four workload configurations (Lite / Standard / Advanced / Publication+) with a recommended primary plan, step-by-step workflow, figure plan, validation strategy, minimal executable version, and publication upgrade path.
Standardmäßig ist der Prompt ausgewählt, der zuerst die Quelle prüft. Sie können zu einem direkten Befehl wechseln oder eine lokale Kopie herunterladen.
Quelldateien prüfen
Lesen Sie SKILL.md und alle von SkillsMP angezeigten Begleitdateien, bevor Sie sich für eine Installation entscheiden.
Mit Codex oder Claude installieren Kopieren Sie diesen Prompt, fügen Sie ihn in Codex, Claude oder einen anderen Assistant ein und lassen Sie die Skill-Seite prüfen und installieren.
Ein direkter Befehl überspringt den Prüf-Prompt. Prüfen Sie die Quelle, bevor Sie ihn ausführen.
Generates complete dual-disease transcriptomic + machine learning research designs from a user-provided disease pair. Use when users want to identify shared DEGs, common hub genes, cross-disease biomarkers, or shared molecular mechanisms between two diseases using public GEO data. Triggers: "shared biomarker study for two diseases", "dual-disease transcriptomic ML paper", "identify common DEGs between disease A and B", "cross-disease hub gene discovery", "shared DEG + PPI + ROC design", "immune infiltration shared biomarker", or "I want to study disease X and Y together". Always outputs four workload configurations (Lite / Standard / Advanced / Publication+) with a recommended primary plan, step-by-step workflow, figure plan, validation strategy, minimal executable version, and publication upgrade path.
Dual-Disease Transcriptomic Machine Learning Research Planner
Generates a complete dual-disease transcriptomic + ML study design from a user-provided disease pair. Always outputs four workload configurations and a recommended primary plan.
Tissue-only filtering: exclude blood/CSF unless disease-appropriate; match tissue type across both diseases
Tissue selection rule: use the tissue most proximal to disease pathology; for metabolic diseases refer to the tissue/tool decision guide
Platform compatibility check: verify GPL IDs match or are cross-compatible before merging
Normalization; batch-awareness without forced merging
Disease vs control group assignment
Fault tolerance — dataset level:
If no GEO dataset exists for one disease: state infeasibility, suggest the closest available proxy phenotype, downgrade to Lite with discovery-only design
If only one dataset is available per disease: downgrade to Lite; clearly state validation ROC is not feasible; provide GEO search strategy for a second cohort
Discovery-cohort ROC + validation-cohort ROC (Standard and above)
Expression validation across cohorts
Fault tolerance — ROC:
If AUC ≈ 0.5 in discovery cohort: do not interpret as biomarker; flag as non-informative; consider mini-signature (3–5 genes) instead of single hub gene
If n < 30 per group: explicitly flag AUC inflation risk; interpret AUC with bootstrap CI; do not generalize
Immune Infiltration (when disease-appropriate per Hard Rule 5)
Hub-prioritization evidence — PPI topology + multi-algorithm consensus (association, not causation)
Biomarker performance evidence — ROC/AUC in discovery + validation cohorts (diagnostic signal, not mechanistic proof)
Immune support — immune landscape differences + gene–immune correlation (associative only; Hard Rule 8)
Single-gene mechanistic support — GSEA pathway themes (hypothesis-generating only; Hard Rule 7)
Step 7: Risk Review
Always include a self-critical section addressing:
Strongest part of the design
Most assumption-dependent part (typically: small cohort ROC inflation; platform differences across datasets)
Most likely false-positive source (hub ranking with few shared DEGs; AUC > 0.9 in n < 50)
Easiest part to overinterpret (immune deconvolution as causal; one hub gene as mechanistic proof)
Most likely reviewer criticisms: small cohorts, no experimental validation, platform heterogeneity, overinterpretation of single biomarker, immune deconvolution limitations, CRC/infectious disease subtype heterogeneity
Revision strategy if first-pass findings fail (broaden DEG threshold, alternate validation cohort, switch to mini-signature)
Step 8: Minimal Executable Version
Public data only, one discovery dataset per disease, DEG + Venn + GO/KEGG, STRING + MCODE + CytoHubba top gene, ROC in discovery cohort, one-page interpretation. 2–4 week timeline. Confirm feasibility against any stated time or dataset constraints before recommending.
When providing R code examples or pipeline frameworks:
EXAMPLE ID convention: All GEO accession numbers in code must carry an inline comment: # EXAMPLE ID — replace with your actual GSE accession before running
Zero-intersection guard: All pipelines must include a feasibility check immediately after DEG intersection:
if(length(shared_genes)==0){
stop("No shared DEGs found. Recovery options: (1) relax logFC to 0.5, (2) use top-500 DEGs per disease, (3) switch to WGCNA co-expression module overlap.")}
Standard package list: GEOquery, limma, clusterProfiler, org.Hs.eg.db, pROC, igraph, STRINGdb, WGCNA. Provide BiocManager::install() calls where needed.
GEO search pattern: To find valid accession IDs, use GEOquery::getGEO("GSEsearch", ...) or direct search at https://www.ncbi.nlm.nih.gov/geo/
Standard R pipeline template:
library(GEOquery); library(limma); library(clusterProfiler); library(pROC)# Load datasets — EXAMPLE IDs: replace before running
gse_disease1 <- getGEO("GSEXXXXX", GSEMatrix =TRUE)[[1]]# EXAMPLE ID
gse_disease2 <- getGEO("GSEXXXXX", GSEMatrix =TRUE)[[1]]# EXAMPLE ID# DEG analysis (repeat for disease2)
design <- model.matrix(~ group, data = pData(gse_disease1))
fit <- eBayes(lmFit(exprs(gse_disease1), design))
deg_d1 <- subset(topTable(fit, coef =2, adjust ="BH", number =Inf),abs(logFC)>1& adj.P.Val <0.05)# Shared DEG intersection with zero-guard
shared_genes <- intersect(rownames(deg_d1), rownames(deg_d2))if(length(shared_genes)==0){
stop("No shared DEGs found. Recovery: relax logFC to 0.5 or use top-500 DEGs per disease.")}# ROC for top hub gene — EXAMPLE: replace 'HUB_GENE' and labels/scores with real data
roc_obj <- roc(response = labels, predictor = expr_scores)
cat("AUC:", auc(roc_obj),"\n")if(auc(roc_obj)<0.70) warning("AUC below 0.70 threshold. Consider mini-signature approach.")
Hard Rules
Never output only one generic plan — always output all four configurations.
Always recommend one primary plan with justification.
Always separate necessary modules from optional modules.
Distinguish shared-expression evidence, biomarker performance evidence, immune support, and mechanistic support — see Step 6.
Do not proceed with immune analysis if the disease pair is not immunologically suited or if deconvolution would be unreliable for the tissue type. Consult references/tissue_and_tool_decisions.md to select the correct tool.
Do not overclaim diagnostic value from ROC in small (n < 30 per group) or unmatched cohorts. Always report bootstrap confidence intervals.
Do not overstate one hub gene as mechanistic proof — label consistently as "biomarker candidate."
Do not treat immune-correlation evidence as causal immune regulation.
If user provides limited detail, infer a reasonable default design and state all assumptions clearly.
Do not produce only a flat methods list or literature summary.
Out-of-scope redirect: If the request involves a single disease only, wet-lab experimental design, clinical trial planning, or non-GEO data types, do not proceed — activate the Input Validation refusal template below.
Input Validation
This skill accepts: a pair of diseases or phenotypes for which the user wants to identify shared transcriptomic signatures, hub genes, or cross-disease biomarkers using publicly available GEO transcriptomic data.
If the request does not involve two diseases for GEO-based transcriptomic comparison — for example, asking to design a study for a single disease only, plan a wet-lab experiment, design a clinical trial, analyze non-transcriptomic omics data (e.g., proteomics, metabolomics), or conduct a systematic literature review — do not proceed with the planning workflow. Instead respond:
"Dual-Disease Transcriptomic ML Planner is designed to generate GEO-based transcriptomic + machine learning study designs for pairs of diseases. Your request appears to be outside this scope. Please provide two diseases to compare, or use a more appropriate skill (e.g., a single-disease transcriptomic skill, an MR planner, or a systematic review skill)."
