Data Analysis

gsva-analysis-and-visualization

Use this skill to run GSVA or ssGSEA pathway-level differential analysis from a bulk expression matrix and a sample group file, then generate a heatmap from the saved GSVA result object.

95100Total Score

Core Capability

100 / 100

Functional Suitability

12 / 12

Reliability

12 / 12

Performance & Context

8 / 8

Agent Usability

16 / 16

Human Usability

8 / 8

Security

12 / 12

Maintainability

12 / 12

Agent-Specific

20 / 20

Medical Task

28 / 29 Passed

97Full GSVA pipeline: bulk matrix vs Tumor/Healthy with KEGG gene sets

5/5

94ssGSEA with MSigDB Hallmarks (H category), top 30 pathways

4/4

90GSVA with 3 samples per group — below recommended minimum

3/4

96Visualization reuse from saved GSVA_list.rda with custom heatmap parameters

4/4

94Full GSVA with C2 REACTOME collection, FDR=0.01, top 50 pathways

4/4

83Single-cell RNA-seq pathway analysis request — out of scope

4/4

91Sample names in group file do not match expression matrix columns

4/4

Veto GatesRequired pass for any deployment consideration

Skill Veto✓ All 4 gates passed

✓

Operational Stability

System remains stable across varied inputs and edge cases

PASS

✓

Structural Consistency

Output structure conforms to expected skill contract format

PASS

✓

Result Determinism

Equivalent inputs produce semantically equivalent outputs

PASS

✓

System Security

No prompt injection, data leakage, or unsafe tool use detected

PASS

Research Veto✅ PASS — Applicable

Dimension	Result	Detail
Scientific Integrity	PASS	No fabricated DOI, PMID, p-values, or pathway enrichment scores across all 7 inputs; all values derived from actual GSVA/limma computation on provided data.
Practice Boundaries	PASS	Clinical diagnosis explicitly excluded in When Not to Use; privacy note for patient-linked data present; no medical recommendations made.
Methodological Ground	PASS	GSVA + limma is the canonical bulk pathway enrichment pipeline; method selection guide correctly advises on GSVA vs ssGSEA for different sample sizes and data characteristics.
Code Usability	PASS	All 9 R modules syntactically valid; dependencies (msigdbr, GSVA, limma, pheatmap) are standard Bioconductor/CRAN packages; three-mode branching logic is correct.

Core Capability100 / 100 — 8 Categories

Functional Suitability

Full coverage of GSVA, ssGSEA, limma differential analysis, heatmap generation, and result object reuse across three execution modes.

12 / 12

100%

Reliability

Hard stops on missing files; visualize mode guards against missing rda; append-only manifests preserve provenance across multiple runs in the same output_dir.

12 / 12

100%

Performance & Context

Progressive disclosure via when-to-read table; references and algorithm details deferred; conditional mode execution avoids loading unnecessary analysis dependencies in visualize-only mode.

8 / 8

100%

Agent Usability

Mode selection guide provides genuine scientific guidance; execution model is explicit; method selection guide warns about sample size requirements; log messages at each step.

16 / 16

100%

Human Usability

Strong trigger keywords (GSVA, ssGSEA, KEGG, MSigDB); privacy note for patient-linked data is a best-practice addition; strict input validation is correct per Scene Override.

8 / 8

100%

Security

No hardcoded secrets; explicit privacy advisory for patient-linked matrices; file path validation; no prompt injection vectors.

12 / 12

100%

Maintainability

9-module R architecture with clean separation; run_tests.R validates both analyze mode and visualize reuse in one pass; test data sourced from public GEO series.

12 / 12

100%

Agent-Specific

Trigger precision excellent with specific NOT-for exclusions and named alternative skills; rda object enables clean downstream composition; append-only manifests support idempotent multi-run workflows.

20 / 20

100%

Core Capability Total100 / 100

Medical TaskExecution Average: 92.1 / 100 — Assertions: 28/29 Passed

Canonical

Full GSVA pipeline: bulk matrix vs Tumor/Healthy with KEGG gene sets

5/5 ✓

Variant A

ssGSEA with MSigDB Hallmarks (H category), top 30 pathways

4/4 ✓

Edge

GSVA with 3 samples per group — below recommended minimum

3/4 ✓

Variant B

Visualization reuse from saved GSVA_list.rda with custom heatmap parameters

4/4 ✓

Stress

Full GSVA with C2 REACTOME collection, FDR=0.01, top 50 pathways

4/4 ✓

Scope Boundary

Single-cell RNA-seq pathway analysis request — out of scope

4/4 ✓

Adversarial

Sample names in group file do not match expression matrix columns

4/4 ✓

Canonical✅ Pass

Full GSVA pipeline: bulk matrix vs Tumor/Healthy with KEGG gene sets

Full mode: load MSigDB KEGG -> GSVA scores -> limma diff analysis -> diff table + score matrices + rda + heatmap. All 8 expected outputs generated.

Basic 40/40|Specialized 57/60|Total 97/100

✅A1Output contains GSVA_diff.csv, GSVA_enrichment_results.csv, GSVA_list.rda, and heatmap PDF

✅A2limma logFC, P.Value, and adj.P.Val columns present in GSVA_diff.csv

✅A3set.seed() applied before GSVA score computation

✅A4No clinical diagnosis or treatment recommendation made

✅A5Privacy note for patient-linked data present in skill documentation

Pass rate: 5 / 5

Variant A✅ Pass

ssGSEA with MSigDB Hallmarks (H category), top 30 pathways

ssGSEA method with Hallmark gene sets; --top_n 30 applied to top pathway subset; GSVA_list.rda saved for downstream reuse.

Basic 39/40|Specialized 55/60|Total 94/100

✅A1ssGSEA method accepted and passed to GSVA computation function

✅A2MSigDB Hallmark category (H) correctly retrieved from msigdbr

✅A3top_n=30 applied to GSVA_enrichment_results_topN.csv pathway subset

✅A4GSVA_list.rda saved for downstream visualization reuse

Pass rate: 4 / 4

Edge✅ Pass

GSVA with 3 samples per group — below recommended minimum

Analysis executes (no hard programmatic block for sample size in GSVA mode); method selection guide warns about minimum 10 samples/group for gsva method but does not emit a runtime warning. Minor gap identified.

Basic 38/40|Specialized 52/60|Total 90/100

✅A1Analysis runs without hard block for insufficient sample count in GSVA mode

✅A2Method selection guide warns users about minimum sample sizes for gsva method

❌A3Runtime log_warn emitted when sample count falls below recommended minimum

✅A4Heatmap generated from top pathways when analysis completes

Pass rate: 3 / 4

Variant B✅ Pass

Visualization reuse from saved GSVA_list.rda with custom heatmap parameters

Visualize mode loads rda without re-running analysis; top_up=10, top_down=10, top_mode=both applied; output appended to existing manifests.

Basic 40/40|Specialized 56/60|Total 96/100

✅A1Visualize mode loads GSVA_list.rda without re-running analysis

✅A2top_up, top_down, top_mode parameters applied to heatmap pathway subset

✅A3SKILL_FILE_NOT_FOUND raised if GSVA_list.rda is missing from output_dir/data/

✅A4Output appended to existing output_manifest.txt and run_record.txt

Pass rate: 4 / 4

Stress✅ Pass

Full GSVA with C2 REACTOME collection, FDR=0.01, top 50 pathways

Large REACTOME collection loaded from msigdbr; FDR=0.01 and top_n=50 applied; timeout parameter available for long runs.

Basic 39/40|Specialized 55/60|Total 94/100

✅A1C2/REACTOME gene sets loaded correctly from msigdbr using --subcategory CP:REACTOME

✅A2FDR threshold 0.01 applied to top pathway selection

✅A3top_n=50 applied to exported pathway score matrix

✅A4Timeout parameter available to bound long computation on large gene set collections

Pass rate: 4 / 4

Scope Boundary✅ Pass

Single-cell RNA-seq pathway analysis request — out of scope

Out-of-scope response triggered; skill correctly declines and names the appropriate alternative workflow. No GSVA execution attempted.

Basic 35/40|Specialized 48/60|Total 83/100

✅A1Single-cell RNA-seq analysis explicitly identified as out of scope in When Not to Use

✅A2Out-of-scope response pattern provided with named alternative skill

✅A3GSVA workflow not executed for single-cell request

✅A4Named alternative skill (sc-clustering) referenced to help user proceed

Pass rate: 4 / 4

Adversarial✅ Pass

Sample names in group file do not match expression matrix columns

SKILL_SAMPLE_MISMATCH raised with actionable message identifying mismatched samples; no analysis proceeds with mismatched data.

Basic 38/40|Specialized 53/60|Total 91/100

✅A1SKILL_SAMPLE_MISMATCH raised when group file samples do not match matrix columns

✅A2Error message identifies the mismatch clearly for agent-driven retry

✅A3No partial GSVA analysis proceeds with mismatched sample data

✅A4SKILL_* error code format enables automated retry with corrected inputs

Pass rate: 4 / 4

Medical Task Total92.1 / 100

Key Strengths

Perfect static score: exemplary SKILL.md with complete documentation of all parameters, outputs, modes, and scientific method selection guidance.
Three-mode execution (analyze/visualize/full) with rda result object reuse is excellent workflow design enabling iterative visualization without expensive re-computation.
Explicit privacy note for patient-linked matrices is a best-practice addition that sets a standard for the skill collection.
Method selection guide (GSVA vs ssGSEA) provides genuine, actionable scientific guidance — not just parameter documentation.