Data Analysis

tf-target-gene-regulatory-network

Analyzes transcription factor (TF) regulatory networks using the Dorothea database. Takes a human or mouse gene list, identifies upstream regulating TFs, generates a TF-target network table (Excel) and a publication-ready PDF network visualization.

94100Total Score

Core Capability

96 / 100

Functional Suitability

12 / 12

Reliability

12 / 12

Performance & Context

8 / 8

Agent Usability

15 / 16

Human Usability

8 / 8

Security

12 / 12

Maintainability

11 / 12

Agent-Specific

18 / 20

Medical Task

33 / 33 Passed

100Human gene list TF network analysis with default parameters

5/5

95Mouse gene list with first-letter-uppercase convention

5/5

95Gene list with no matching TF-target relationships in Dorothea

5/5

95Gene file input mode (--gene_file instead of --gene)

5/5

92Large gene list (>500 genes) with local Dorothea RDS database

5/5

85Pathway enrichment analysis request

4/4

85Mixed-species gene list (human and mouse symbols together)

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 TF-target relationships; all interactions sourced from Dorothea curated database with documented confidence levels A/B/C.
Practice Boundaries	PASS	No clinical diagnostic conclusions; correctly scoped as a TF regulatory network research tool with no causal inference beyond curated Dorothea relationships.
Methodological Ground	PASS	Dorothea database queries with confidence-level filtering (A/B/C) is a standard correct method for TF enrichment analysis; SKILL.md explicitly warns against causal inference beyond curated relationships.
Code Usability	PASS	main.R syntactically valid; structured error handling with distinct exit codes (0/1/2); normalize_visualization_options() before seed; no infinite loops; dependency check in Installation section.

Core Capability96 / 100 — 8 Categories

Functional Suitability

Full pipeline: Dorothea DB load (local or package), TF identification, network table (Excel with edges+nodes worksheets), PDF visualization, session_info; human and mouse supported; local RDS for offline use.

12 / 12

100%

Reliability

Six SKILL_* codes plus three-tier exit status (0=success, 1=error, 2=SKILL_EMPTY_RESULTS); distinct exit 2 for no-TF-match case allows agent to differentiate no-result from failure.

12 / 12

100%

Performance & Context

Progressive reference loading; visualization-parameters.md referenced only when needed; local DB path auto-search prevents redundant downloads.

8 / 8

100%

Agent Usability

Clear When to Use/Not to Use, Installation, four-step Workflow, Arguments table; legacy Chinese alias support documented but creates minor cognitive load; minor: no post-run agent output checklist.

15 / 16

94%

Human Usability

Precise trigger language; gene naming convention (case-sensitive per species) clearly documented; strict species validation is correct per Scene Override.

8 / 8

100%

Security

No credentials; gene list parsed from string or file; no exec/eval of user strings; local DB path validated; species limited to human/mouse.

12 / 12

100%

Maintainability

Four well-separated scripts plus database-get.R; visualization isolated in visualization.R; no formal test runner script, only manual test commands.

11 / 12

92%

Agent-Specific

Excellent trigger precision and idempotency; two-way gene input (--gene or --gene_file) improves composability; tf.Rdata output is opaque to downstream tools reducing composability score; composability not explicitly documented.

18 / 20

90%

Core Capability Total96 / 100

Medical TaskExecution Average: 92.4 / 100 — Assertions: 33/33 Passed

100

Canonical

Human gene list TF network analysis with default parameters

5/5 ✓

Variant A

Mouse gene list with first-letter-uppercase convention

5/5 ✓

Edge

Gene list with no matching TF-target relationships in Dorothea

5/5 ✓

Variant B

Gene file input mode (--gene_file instead of --gene)

5/5 ✓

Stress

Large gene list (>500 genes) with local Dorothea RDS database

5/5 ✓

Scope Boundary

Pathway enrichment analysis request

4/4 ✓

Adversarial

Mixed-species gene list (human and mouse symbols together)

4/4 ✓

100

Canonical✅ Pass

Human gene list TF network analysis with default parameters

gene=TP53,MYC,EGFR, species=human, seed=42; Dorothea DB loaded (local .rds or package); TF-target pairs identified; tf_network.xlsx, TF_Target_Filtered_Core_human.xlsx, TF_Network_Plot.pdf, session_info.txt produced.

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

✅A1Output produces tf_network.xlsx with edges and nodes worksheets

✅A2Dorothea confidence levels A/B/C applied to filter TF-target interactions

✅A3Random seed set before visualization ensuring reproducible layout

✅A4Output does not infer causal direction beyond curated Dorothea relationships

✅A5session_info.txt saved for reproducibility audit

Pass rate: 5 / 5

Variant A✅ Pass

Mouse gene list with first-letter-uppercase convention

gene=Tp53,Myc,Egfr, species=mouse; dorothea_mm.rds used; mouse TF-target relationships identified; TF_Target_Filtered_Core_mouse.xlsx produced.

Basic 38/40|Specialized 57/60|Total 95/100

✅A1Mouse gene symbols use first-letter-uppercase convention (Tp53 not TP53)

✅A2Mouse Dorothea database (dorothea_mm.rds) loaded for species=mouse

✅A3Output file named TF_Target_Filtered_Core_mouse.xlsx for mouse species

✅A4Fixed seed ensures reproducible mouse network layout

✅A5Scope maintained — mouse TF analysis is within stated species support

Pass rate: 5 / 5

Edge✅ Pass

Gene list with no matching TF-target relationships in Dorothea

SKILL_EMPTY_RESULTS raised; exit status 2 (distinct from error exit 1); allows agent to differentiate no-match from failure.

Basic 40/40|Specialized 55/60|Total 95/100

✅A1SKILL_EMPTY_RESULTS raised when no TF-target relationships found

✅A2Exit status 2 is distinct from exit status 1 (execution error)

✅A3Error message suggests checking gene symbols and species

✅A4No fabricated TF-target relationships generated for non-matching genes

✅A5Scope not exceeded — correct hard stop rather than incorrect empty output

Pass rate: 5 / 5

Variant B✅ Pass

Gene file input mode (--gene_file instead of --gene)

gene_file=genes.txt with one gene per line; parsed correctly; same output as --gene mode.

Basic 38/40|Specialized 57/60|Total 95/100

✅A1Gene file input (--gene_file) correctly parses one-per-line or comma-separated genes

✅A2Either --gene or --gene_file must be provided — both-absent raises error

✅A3Output files identical between --gene and --gene_file modes

✅A4Fixed seed ensures reproducible results regardless of input method

✅A5Scope maintained — file input is a documented alternative input method

Pass rate: 5 / 5

Stress✅ Pass

Large gene list (>500 genes) with local Dorothea RDS database

Performance advisory for >500 genes documented; local --db_path recommended; no timeout parameter available — agent should monitor manually.

Basic 38/40|Specialized 54/60|Total 92/100

✅A1Local --db_path recommended for large gene lists to avoid slow network queries

✅A2Auto-search paths for local database files documented (4 paths checked)

✅A3No timeout_seconds parameter available — agent advised to monitor progress

✅A4Fixed seed ensures reproducible network layout for large gene sets

✅A5Scope maintained — large gene list analysis is within stated capability

Pass rate: 5 / 5

Scope Boundary✅ Pass

Pathway enrichment analysis request

Input Validation guard fires; pathway enrichment explicitly excluded in When Not to Use.

Basic 35/40|Specialized 50/60|Total 85/100

✅A1Pathway enrichment request refused by Input Validation guard

✅A2Refusal message clearly states TF regulatory analysis scope

✅A3No incorrect pathway enrichment computation attempted

✅A4No fabricated pathway enrichment results produced

Pass rate: 4 / 4

Adversarial✅ Pass

Mixed-species gene list (human and mouse symbols together)

SKILL.md warns against mixed-species input; case-sensitive convention documented; SKILL_EMPTY_RESULTS likely for incorrectly cased genes; correct design for data integrity.

Basic 35/40|Specialized 50/60|Total 85/100

✅A1Mixed-species input warning documented — user advised to normalize symbols first

✅A2Case-sensitive gene naming convention enforced — TP53 (human) vs Tp53 (mouse)

✅A3SKILL_EMPTY_RESULTS correctly raised for genes with no Dorothea matches due to case mismatch

✅A4Scope maintained — no auto-normalization of mixed-species input attempted

Pass rate: 4 / 4

Medical Task Total92.4 / 100

Key Strengths

Three-tier exit code design (0=success, 1=error, 2=SKILL_EMPTY_RESULTS) enables precise agent-level differentiation between failure and no-TF-match outcomes
Local Dorothea RDS database support with 4-path auto-search enables offline air-gapped execution and substantially faster lookups for large gene lists
Both --gene (inline) and --gene_file (file) input modes supported, improving composability for agents working with existing gene list files
Explicit warning against causal inference beyond curated Dorothea relationships maintains scientific integrity at the SKILL.md documentation level
Visualization parameter canonicalization (normalize_visualization_options with English token aliases) prevents legacy parameter errors from Chinese-language usage