mims-harvard/tooluniverse-multiomic-disease-characterization

Multi-Omics Disease Characterization Pipeline

Characterize diseases across multiple molecular layers (genomics, transcriptomics, proteomics, pathways) to provide systems-level understanding of disease mechanisms, identify therapeutic opportunities, and discover biomarker candidates.

KEY PRINCIPLES:

1. Report-first approach - Create report file FIRST, then populate progressively
2. Disease disambiguation FIRST - Resolve all identifiers before omics analysis
3. Layer-by-layer analysis - Systematically cover all omics layers
4. Cross-layer integration - Identify genes/targets appearing in multiple layers
5. Evidence grading - Grade all evidence as T1 (human/clinical) to T4 (computational)
6. Tissue context - Emphasize disease-relevant tissues/organs
7. Quantitative scoring - Multi-Omics Confidence Score (0-100)
8. Druggable focus - Prioritize targets with therapeutic potential
9. Biomarker identification - Highlight diagnostic/prognostic markers
10. Mechanistic synthesis - Generate testable hypotheses
11. Source references - Every statement must cite tool/database
12. Completeness checklist - Mandatory section showing analysis coverage
13. English-first queries - Always use English terms in tool calls. Respond in user's language

Multi-omics disease characterization asks: what molecular layers are dysregulated? Genomic mutations → transcriptomic changes → proteomic effects → metabolomic consequences. Concordance across layers strengthens the finding. Discordance reveals regulatory complexity.

LOOK UP, DON'T GUESS

When uncertain about any scientific fact, SEARCH databases first rather than reasoning from memory. A database-verified answer is always more reliable than a guess.

---

COMPUTE, DON'T DESCRIBE

When analysis requires computation (statistics, data processing, scoring, enrichment), write and run Python code via Bash. Don't describe what you would do — execute it and report actual results. Use ToolUniverse tools to retrieve data, then Python (pandas, scipy, statsmodels, matplotlib) to analyze it.

When to Use This Skill

Apply when users:

• Ask about disease mechanisms across omics layers
• Need multi-omics characterization of a disease
• Want to understand disease at the systems biology level
• Ask "What pathways/genes/proteins are involved in [disease]?"
• Need biomarker discovery for a disease
• Want to identify druggable targets from disease profiling
• Ask for integrated genomics + transcriptomics + proteomics analysis
• Need cross-layer concordance analysis
• Ask about disease network biology / hub genes

NOT for (use other skills instead):

• Single gene/target validation -> Use tooluniverse-drug-target-validation
• Drug safety profiling -> Use tooluniverse-adverse-event-detection
• General disease overview -> Use tooluniverse-disease-research
• Variant interpretation -> Use tooluniverse-variant-interpretation
• GWAS-specific analysis -> Use tooluniverse-gwas-* skills
• Pathway-only analysis -> Use tooluniverse-systems-biology

---

Input Parameters

| Parameter | Required | Description | Example | |-----------|----------|-------------|---------| | disease | Yes | Disease name, OMIM ID, EFO ID, or MONDO ID | Alzheimer disease, MONDO_0004975 | | tissue | No | Tissue/organ of interest | brain, liver, blood | | focus_layers | No | Specific omics layers to emphasize | genomics, transcriptomics, pathways |

---

Pipeline Overview

The pipeline runs 9 phases sequentially. Each phase uses specific tools documented in detail in tool-reference.md.

Phase 0: Disease Disambiguation (ALWAYS FIRST)

Resolve disease to standard identifiers (MONDO/EFO) for all downstream queries.

• Primary tool: OpenTargets_get_disease_id_description_by_name
• Get description, synonyms, therapeutic areas, disease hierarchy, cross-references
• CRITICAL: Disease IDs use underscore format (e.g., MONDO_0004975), NOT colon
• If ambiguous, present top 3-5 options and ask user to select

Phase 1: Genomics Layer

Identify genetic variants, GWAS associations, and genetically implicated genes.

• Tools: gwas_search_associations (use efo_id for precision, not free-text disease_trait), gwas_get_snps_for_gene, ClinVar, OpenTargets associated targets
• gnomad_get_gene_constraints — gene constraint metrics (pLI, oe_lof) to interpret whether LoF variants are tolerated vs. haploinsufficient
• Get top 10-15 genes with genetic evidence scores; track Ensembl IDs for downstream phases

Phase 2: Transcriptomics Layer

Identify differentially expressed genes, tissue-specific expression, and expression-based biomarkers.

• GTEx_get_expression_summary — baseline expression across 54 tissues (accepts gene_symbol directly)
• Tools: Expression Atlas, HPA (tissue expression), EuropePMC scores
• Check expression in disease-relevant tissues for top genes from Phase 1

Phase 3: Proteomics & Interaction Layer

Map protein-protein interactions, identify hub genes, and characterize interaction networks.

• UniProt_get_function_by_accession — protein function narrative (essential for mechanistic context)
• Tools: STRING_get_network (param: identifiers, species=9606), intact_get_interactions, HumanBase
• Build PPI network from top 15-20 genes; identify hub genes by degree centrality

Phase 4: Pathway & Network Layer

Identify enriched biological pathways and cross-pathway connections.

• ReactomeAnalysis_pathway_enrichment — identifiers are newline-separated (\n), NOT space-separated
• enrichr_gene_enrichment_analysis — param: gene_list (array), libs (array). NOTE: data field is a JSON string that needs parsing
• kegg_search_pathway — pathway keyword search

Phase 5: Gene Ontology & Functional Annotation

Characterize biological processes, molecular functions, and cellular components.

• Tools: Enrichr (GO libraries), QuickGO, GO annotations, OpenTargets GO
• Run GO enrichment for all 3 aspects (BP, MF, CC)

Phase 6: Therapeutic Landscape

Map approved drugs, druggable targets, repurposing opportunities, and clinical trials.

• DGIdb_get_drug_gene_interactions — drug interactions by gene (param: genes as array). Often more comprehensive than OpenTargets for drug-gene data.
• OpenTargets drugs/tractability (use EFO IDs like EFO_0000384 for Crohn's, not MONDO — MONDO IDs may return null for drug queries)
• search_clinical_trials — query_term is REQUIRED

Phase 7: Multi-Omics Integration

Integrate findings across all layers. See integration-scoring.md for full details.

• Cross-layer gene concordance: count layers per gene, score multi-layer hub genes
• Direction concordance: genetics + expression agreement
• Biomarker identification: diagnostic, prognostic, predictive
• Mechanistic hypothesis generation

Phase 8: Report Finalization

Write executive summary, calculate confidence score, verify completeness.

• See integration-scoring.md for quality checklist and scoring formula

---

Key Tool Parameter Notes

These are the most common parameter pitfalls:

• OpenTargets disease IDs: underscore format (MONDO_0004975), NOT colon
• STRING protein_ids: must be array (['APOE']), not string
• enrichr libs: must be array (['KEGG_2021_Human'])
• HPA_get_rna_expression_by_source: ALL 3 params required (gene_name, source_type, source_name)
• humanbase_ppi_analysis: ALL params required (gene_list, tissue, max_node, interaction, string_mode)
• expression_atlas_disease_target_score: pageSize is REQUIRED
• search_clinical_trials: query_term is REQUIRED even if condition is provided

For full tool parameters and per-phase workflows, see tool-reference.md.

---

Reference Files

All detailed content is in reference files in this directory:

| File | Contents | |------|----------| | tool-reference.md | Full tool parameters, inputs/outputs, per-phase workflows, quick reference table | | report-template.md | Complete report markdown template with all sections and checklists | | integration-scoring.md | Confidence score formula (0-100), evidence grading (T1-T4), integration procedures, quality checklist | | response-formats.md | Verified JSON response structures for key tools | | use-patterns.md | Common use patterns, edge case handling, fallback strategies |