Precision Medicine Patient Stratification
Transform patient genomic and clinical profiles into actionable risk stratification, treatment recommendations, and personalized therapeutic strategies.
Reasoning Before Searching
Stratification means splitting patients into groups that respond differently to a treatment or have different prognoses. Ask these questions before running any tools:
- What molecular feature predicts response? Candidates: somatic mutation (e.g., EGFR L858R), germline variant (e.g., BRCA1 LoF), expression level (e.g., HER2 overexpression), germline pharmacogenomic variant (e.g., CYP2C19 PM), or composite biomarker (e.g., TMB-H + MSI-H).
- Is the predictive feature actionable? Knowing it must change treatment — either the drug choice, dose, or monitoring plan. A variant with prognostic value but no therapeutic consequence is not a stratification biomarker.
- What is the evidence level for the stratifier? FDA-approved companion diagnostic (T1) vs. exploratory (T4) changes how much weight to place on the finding.
Route to the correct Phase 3 path BEFORE running Phase 2 tools — cancer, metabolic, CVD, rare disease, and autoimmune pipelines require different stratifiers.
LOOK UP DON'T GUESS: Never assume a variant is pathogenic, never assume a gene is relevant to a disease, never assign metabolizer status without PharmGKB or CPIC evidence.
KEY PRINCIPLES:
- Report-first - Create report file FIRST, then populate progressively
- Disease-specific logic - Cancer vs metabolic vs rare disease pipelines diverge at Phase 3
- Multi-level integration - Germline + somatic + expression + clinical data layers
- Evidence-graded - Every finding has an evidence tier (T1-T4)
- Quantitative output - Precision Medicine Risk Score (0-100)
- Source-referenced - Every statement cites the tool/database source
- English-first queries - Always use English terms in tool calls
Reference files (same directory):
TOOLS_REFERENCE.md - Tool parameters, response formats, phase-by-phase tool lists
SCORING_REFERENCE.md - Scoring matrices, risk tiers, pathogenicity tables, PGx tables
REPORT_TEMPLATE.md - Output report template, treatment algorithms, completeness requirements
EXAMPLES.md - Six worked examples (cancer, metabolic, NSCLC, CVD, rare, neuro)
QUICK_START.md - Sample prompts and output summary
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
Apply when user asks about patient risk stratification, treatment selection, prognosis prediction, or personalized therapeutic strategy for any disease with genomic/clinical data.
NOT for (use other skills instead):
- Single variant interpretation ->
tooluniverse-variant-interpretation
- Immunotherapy-specific prediction ->
tooluniverse-immunotherapy-response-prediction
- Drug safety profiling only ->
tooluniverse-adverse-event-detection
- Target validation ->
tooluniverse-drug-target-validation
- Clinical trial search only ->
tooluniverse-clinical-trial-matching
- Drug-drug interaction only ->
tooluniverse-drug-drug-interaction
- PRS calculation only ->
tooluniverse-polygenic-risk-score
Input Parsing
Required
- Disease/condition: Free-text disease name
- At least one of: Germline variants, somatic mutations, gene list, or clinical biomarkers
Optional (improves stratification)
- Age, sex, ethnicity, disease stage, comorbidities, prior treatments, family history
- Current medications (for DDI and PGx), stratification goal
Disease Type Classification
Classify into one category (determines Phase 3 routing):
| Category |
Examples |
| CANCER |
Breast, lung, colorectal, melanoma |
| METABOLIC |
Type 2 diabetes, obesity, NAFLD |
| CARDIOVASCULAR |
CAD, heart failure, AF |
| NEUROLOGICAL |
Alzheimer, Parkinson, epilepsy |
| RARE/MONOGENIC |
Marfan, CF, sickle cell, Huntington |
| AUTOIMMUNE |
RA, lupus, MS, Crohn's |
Critical Tool Parameter Notes
See TOOLS_REFERENCE.md for full details. Key gotchas:
- MyGene_query_genes: param is
query (NOT q)
- EnsemblVEP_annotate_rsid: param is
variant_id (NOT rsid)
- ensembl_lookup_gene: REQUIRES
species='homo_sapiens'
- DrugBank tools: ALL require 4 params:
query, case_sensitive, exact_match, limit
- cBioPortal_get_mutations:
gene_list is a STRING (space-separated), not array
- PubMed_search_articles: Returns a plain list of dicts, NOT
{articles: [...]}
- fda_pharmacogenomic_biomarkers: Use
limit=1000 for all results
- gnomAD: May return "Service overloaded" - skip gracefully
- OpenTargets: Always nested
{data: {entity: {field: ...}}} structure
Workflow Overview
Phase 1: Disease Disambiguation & Profile Standardization
Phase 2: Genetic Risk Assessment
Phase 3: Disease-Specific Molecular Stratification (routes by disease type)
Phase 4: Pharmacogenomic Profiling
Phase 5: Comorbidity & Drug Interaction Risk
Phase 6: Molecular Pathway Analysis
Phase 7: Clinical Evidence & Guidelines
Phase 8: Clinical Trial Matching
Phase 9: Integrated Scoring & Recommendations
Phase 1: Disease Disambiguation & Profile Standardization
- Resolve disease to EFO ID using
OpenTargets_get_disease_id_description_by_name
- Classify disease type (CANCER/METABOLIC/CVD/NEUROLOGICAL/RARE/AUTOIMMUNE)
- Parse genomic data into structured format (gene, variant, type)
- Resolve gene IDs using
MyGene_query_genes to get Ensembl/Entrez IDs
Phase 2: Genetic Risk Assessment
- Germline variant pathogenicity:
ClinVar_search_variants, EnsemblVEP_annotate_rsid/_hgvs
- Gene-disease association:
OpenTargets_target_disease_evidence
- GWAS polygenic risk:
gwas_get_associations_for_trait, OpenTargets_search_gwas_studies_by_disease
- Population frequency:
gnomad_get_variant
- Gene constraint:
gnomad_get_gene_constraints (pLI, LOEUF scores)
Scoring: See SCORING_REFERENCE.md for genetic risk score component (0-35 points).
Phase 3: Disease-Specific Molecular Stratification
CANCER PATH
- Molecular subtyping:
cBioPortal_get_mutations, HPA_get_cancer_prognostics_by_gene
- TMB/MSI/HRD:
fda_pharmacogenomic_biomarkers for FDA cutoffs
- Prognostic stratification: Combine stage + molecular features
METABOLIC PATH
- Genetic risk integration:
GWAS_search_associations_by_gene, OpenTargets_target_disease_evidence
- Complication risk: Based on HbA1c, duration, existing complications
CVD PATH
- FH gene check:
ClinVar_search_variants for LDLR, APOB, PCSK9
- Statin PGx:
PharmGKB_get_clinical_annotations for SLCO1B1
RARE DISEASE PATH
- Causal variant identification:
ClinVar_search_variants
- Genotype-phenotype:
UniProt_get_disease_variants_by_accession
Scoring: See SCORING_REFERENCE.md for disease-specific tables.
Phase 4: Pharmacogenomic Profiling
- Drug-metabolizing enzymes:
PharmGKB_get_clinical_annotations, PharmGKB_get_dosing_guidelines
- FDA PGx biomarkers:
fda_pharmacogenomic_biomarkers (use limit=1000)
- Treatment-specific PGx:
PharmGKB_get_drug_details
Scoring: See SCORING_REFERENCE.md for PGx risk score (0-10 points).
Phase 5: Comorbidity & Drug Interaction Risk
- Disease overlap:
OpenTargets_get_associated_targets_by_disease_efoId
- DDI check:
drugbank_get_drug_interactions_by_drug_name_or_id, FDA_get_drug_interactions_by_drug_name
- PGx-amplified DDI: If PM genotype + CYP inhibitor, flag compounded risk
Phase 6: Molecular Pathway Analysis
- Pathway enrichment:
enrichr_gene_enrichment_analysis (libs: KEGG_2021_Human, Reactome_2022, GO_Biological_Process_2023)
- Reactome mapping:
ReactomeAnalysis_pathway_enrichment, Reactome_map_uniprot_to_pathways
- Network analysis:
STRING_get_interaction_partners, STRING_functional_enrichment
- Druggable targets:
OpenTargets_get_target_tractability_by_ensemblID
Phase 7: Clinical Evidence & Guidelines
- Guidelines search:
PubMed_Guidelines_Search (fallback: PubMed_search_articles)
- FDA-approved therapies:
OpenTargets_get_associated_drugs_by_disease_efoId, FDA_get_indications_by_drug_name
- Biomarker-drug evidence:
civic_search_evidence_items, civic_search_assertions
Phase 8: Clinical Trial Matching
- Biomarker-driven trials:
search_clinical_trials with condition + intervention
- Precision medicine trials:
search_clinical_trials for basket/umbrella trials
Phase 9: Integrated Scoring & Recommendations
Score Components (total 0-100)
- Genetic Risk (0-35): Pathogenicity + gene-disease association + PRS
- Clinical Risk (0-30): Stage/biomarkers/comorbidities
- Molecular Features (0-25): Driver mutations, subtypes, actionable targets
- Pharmacogenomic Risk (0-10): Metabolizer status, HLA alleles
Risk Tiers
| Score |
Tier |
Management |
| 75-100 |
VERY HIGH |
Intensive treatment, subspecialty referral, clinical trial |
| 50-74 |
HIGH |
Aggressive treatment, close monitoring |
| 25-49 |
INTERMEDIATE |
Standard guideline-based care, PGx-guided dosing |
| 0-24 |
LOW |
Surveillance, prevention, risk factor modification |
Output
Generate report per REPORT_TEMPLATE.md. See SCORING_REFERENCE.md for detailed scoring matrices.
Common Use Patterns
See EXAMPLES.md for six detailed worked examples:
- Cancer + actionable mutation: Breast cancer, BRCA1, ER+/HER2- -> Score ~55-65 (HIGH)
- Metabolic + PGx concern: T2D, CYP2C19 PM on clopidogrel -> Score ~55-65 (HIGH)
- NSCLC comprehensive: EGFR L858R, TMB 25, PD-L1 80% -> Score ~75-85 (VERY HIGH)
- CVD risk: LDL 190, SLCO1B1*5, family hx MI -> Score ~50-60 (HIGH)
- Rare disease: Marfan, FBN1 variant -> Score ~55-65 (HIGH)
- Neurological risk: APOE e4/e4, family hx Alzheimer's -> Score ~60-72 (HIGH)
1---23name: tooluniverse-precision-medicine-stratification-24description: "Patient stratification for precision medicine — integrate genomic, clinical, and therapeutic data to split patients into responder/non-responder groups, risk tiers, or treatment-decision groups. Use for stratification-by-biomarker, treatment-selection logic, and personalized therapeutic strategy reports per patient subgroup."5metadata:6 source: https://skillmd.com/skills/mims-harvard/tooluniverse-precision-medicine-stratification-27---89# Precision Medicine Patient Stratification1011Transform patient genomic and clinical profiles into actionable risk stratification, treatment recommendations, and personalized therapeutic strategies.1213## Reasoning Before Searching1415Stratification means splitting patients into groups that respond differently to a treatment or have different prognoses. Ask these questions before running any tools:16171. **What molecular feature predicts response?** Candidates: somatic mutation (e.g., EGFR L858R), germline variant (e.g., BRCA1 LoF), expression level (e.g., HER2 overexpression), germline pharmacogenomic variant (e.g., CYP2C19 PM), or composite biomarker (e.g., TMB-H + MSI-H).182. **Is the predictive feature actionable?** Knowing it must change treatment — either the drug choice, dose, or monitoring plan. A variant with prognostic value but no therapeutic consequence is not a stratification biomarker.193. **What is the evidence level for the stratifier?** FDA-approved companion diagnostic (T1) vs. exploratory (T4) changes how much weight to place on the finding.2021Route to the correct Phase 3 path BEFORE running Phase 2 tools — cancer, metabolic, CVD, rare disease, and autoimmune pipelines require different stratifiers.2223**LOOK UP DON'T GUESS**: Never assume a variant is pathogenic, never assume a gene is relevant to a disease, never assign metabolizer status without PharmGKB or CPIC evidence.2425**KEY PRINCIPLES**:261. **Report-first** - Create report file FIRST, then populate progressively272. **Disease-specific logic** - Cancer vs metabolic vs rare disease pipelines diverge at Phase 3283. **Multi-level integration** - Germline + somatic + expression + clinical data layers294. **Evidence-graded** - Every finding has an evidence tier (T1-T4)305. **Quantitative output** - Precision Medicine Risk Score (0-100)316. **Source-referenced** - Every statement cites the tool/database source327. **English-first queries** - Always use English terms in tool calls3334**Reference files** (same directory):35- `TOOLS_REFERENCE.md` - Tool parameters, response formats, phase-by-phase tool lists36- `SCORING_REFERENCE.md` - Scoring matrices, risk tiers, pathogenicity tables, PGx tables37- `REPORT_TEMPLATE.md` - Output report template, treatment algorithms, completeness requirements38- `EXAMPLES.md` - Six worked examples (cancer, metabolic, NSCLC, CVD, rare, neuro)39- `QUICK_START.md` - Sample prompts and output summary4041---4243## COMPUTE, DON'T DESCRIBE44When 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.4546## When to Use4748Apply when user asks about patient risk stratification, treatment selection, prognosis prediction, or personalized therapeutic strategy for any disease with genomic/clinical data.4950**NOT for** (use other skills instead):51- Single variant interpretation -> `tooluniverse-variant-interpretation`52- Immunotherapy-specific prediction -> `tooluniverse-immunotherapy-response-prediction`53- Drug safety profiling only -> `tooluniverse-adverse-event-detection`54- Target validation -> `tooluniverse-drug-target-validation`55- Clinical trial search only -> `tooluniverse-clinical-trial-matching`56- Drug-drug interaction only -> `tooluniverse-drug-drug-interaction`57- PRS calculation only -> `tooluniverse-polygenic-risk-score`5859---6061## Input Parsing6263### Required64- **Disease/condition**: Free-text disease name65- **At least one of**: Germline variants, somatic mutations, gene list, or clinical biomarkers6667### Optional (improves stratification)68- Age, sex, ethnicity, disease stage, comorbidities, prior treatments, family history69- Current medications (for DDI and PGx), stratification goal7071### Disease Type Classification7273Classify into one category (determines Phase 3 routing):7475| Category | Examples |76|----------|----------|77| **CANCER** | Breast, lung, colorectal, melanoma |78| **METABOLIC** | Type 2 diabetes, obesity, NAFLD |79| **CARDIOVASCULAR** | CAD, heart failure, AF |80| **NEUROLOGICAL** | Alzheimer, Parkinson, epilepsy |81| **RARE/MONOGENIC** | Marfan, CF, sickle cell, Huntington |82| **AUTOIMMUNE** | RA, lupus, MS, Crohn's |8384---8586## Critical Tool Parameter Notes8788See `TOOLS_REFERENCE.md` for full details. Key gotchas:8990- **MyGene_query_genes**: param is `query` (NOT `q`)91- **EnsemblVEP_annotate_rsid**: param is `variant_id` (NOT `rsid`)92- **ensembl_lookup_gene**: REQUIRES `species='homo_sapiens'`93- **DrugBank tools**: ALL require 4 params: `query`, `case_sensitive`, `exact_match`, `limit`94- **cBioPortal_get_mutations**: `gene_list` is a STRING (space-separated), not array95- **PubMed_search_articles**: Returns a plain list of dicts, NOT `{articles: [...]}`96- **fda_pharmacogenomic_biomarkers**: Use `limit=1000` for all results97- **gnomAD**: May return "Service overloaded" - skip gracefully98- **OpenTargets**: Always nested `{data: {entity: {field: ...}}}` structure99100---101102## Workflow Overview103104```105Phase 1: Disease Disambiguation & Profile Standardization106Phase 2: Genetic Risk Assessment107Phase 3: Disease-Specific Molecular Stratification (routes by disease type)108Phase 4: Pharmacogenomic Profiling109Phase 5: Comorbidity & Drug Interaction Risk110Phase 6: Molecular Pathway Analysis111Phase 7: Clinical Evidence & Guidelines112Phase 8: Clinical Trial Matching113Phase 9: Integrated Scoring & Recommendations114```115116---117118## Phase 1: Disease Disambiguation & Profile Standardization1191201. **Resolve disease to EFO ID** using `OpenTargets_get_disease_id_description_by_name`1212. **Classify disease type** (CANCER/METABOLIC/CVD/NEUROLOGICAL/RARE/AUTOIMMUNE)1223. **Parse genomic data** into structured format (gene, variant, type)1234. **Resolve gene IDs** using `MyGene_query_genes` to get Ensembl/Entrez IDs124125## Phase 2: Genetic Risk Assessment1261271. **Germline variant pathogenicity**: `ClinVar_search_variants`, `EnsemblVEP_annotate_rsid`/`_hgvs`1282. **Gene-disease association**: `OpenTargets_target_disease_evidence`1293. **GWAS polygenic risk**: `gwas_get_associations_for_trait`, `OpenTargets_search_gwas_studies_by_disease`1304. **Population frequency**: `gnomad_get_variant`1315. **Gene constraint**: `gnomad_get_gene_constraints` (pLI, LOEUF scores)132133Scoring: See `SCORING_REFERENCE.md` for genetic risk score component (0-35 points).134135## Phase 3: Disease-Specific Molecular Stratification136137### CANCER PATH1381. **Molecular subtyping**: `cBioPortal_get_mutations`, `HPA_get_cancer_prognostics_by_gene`1392. **TMB/MSI/HRD**: `fda_pharmacogenomic_biomarkers` for FDA cutoffs1403. **Prognostic stratification**: Combine stage + molecular features141142### METABOLIC PATH1431. **Genetic risk integration**: `GWAS_search_associations_by_gene`, `OpenTargets_target_disease_evidence`1442. **Complication risk**: Based on HbA1c, duration, existing complications145146### CVD PATH1471. **FH gene check**: `ClinVar_search_variants` for LDLR, APOB, PCSK91482. **Statin PGx**: `PharmGKB_get_clinical_annotations` for SLCO1B1149150### RARE DISEASE PATH1511. **Causal variant identification**: `ClinVar_search_variants`1522. **Genotype-phenotype**: `UniProt_get_disease_variants_by_accession`153154Scoring: See `SCORING_REFERENCE.md` for disease-specific tables.155156## Phase 4: Pharmacogenomic Profiling1571581. **Drug-metabolizing enzymes**: `PharmGKB_get_clinical_annotations`, `PharmGKB_get_dosing_guidelines`1592. **FDA PGx biomarkers**: `fda_pharmacogenomic_biomarkers` (use `limit=1000`)1603. **Treatment-specific PGx**: `PharmGKB_get_drug_details`161162Scoring: See `SCORING_REFERENCE.md` for PGx risk score (0-10 points).163164## Phase 5: Comorbidity & Drug Interaction Risk1651661. **Disease overlap**: `OpenTargets_get_associated_targets_by_disease_efoId`1672. **DDI check**: `drugbank_get_drug_interactions_by_drug_name_or_id`, `FDA_get_drug_interactions_by_drug_name`1683. **PGx-amplified DDI**: If PM genotype + CYP inhibitor, flag compounded risk169170## Phase 6: Molecular Pathway Analysis1711721. **Pathway enrichment**: `enrichr_gene_enrichment_analysis` (libs: `KEGG_2021_Human`, `Reactome_2022`, `GO_Biological_Process_2023`)1732. **Reactome mapping**: `ReactomeAnalysis_pathway_enrichment`, `Reactome_map_uniprot_to_pathways`1743. **Network analysis**: `STRING_get_interaction_partners`, `STRING_functional_enrichment`1754. **Druggable targets**: `OpenTargets_get_target_tractability_by_ensemblID`176177## Phase 7: Clinical Evidence & Guidelines1781791. **Guidelines search**: `PubMed_Guidelines_Search` (fallback: `PubMed_search_articles`)1802. **FDA-approved therapies**: `OpenTargets_get_associated_drugs_by_disease_efoId`, `FDA_get_indications_by_drug_name`1813. **Biomarker-drug evidence**: `civic_search_evidence_items`, `civic_search_assertions`182183## Phase 8: Clinical Trial Matching1841851. **Biomarker-driven trials**: `search_clinical_trials` with condition + intervention1862. **Precision medicine trials**: `search_clinical_trials` for basket/umbrella trials187188## Phase 9: Integrated Scoring & Recommendations189190### Score Components (total 0-100)191- **Genetic Risk** (0-35): Pathogenicity + gene-disease association + PRS192- **Clinical Risk** (0-30): Stage/biomarkers/comorbidities193- **Molecular Features** (0-25): Driver mutations, subtypes, actionable targets194- **Pharmacogenomic Risk** (0-10): Metabolizer status, HLA alleles195196### Risk Tiers197| Score | Tier | Management |198|-------|------|------------|199| 75-100 | VERY HIGH | Intensive treatment, subspecialty referral, clinical trial |200| 50-74 | HIGH | Aggressive treatment, close monitoring |201| 25-49 | INTERMEDIATE | Standard guideline-based care, PGx-guided dosing |202| 0-24 | LOW | Surveillance, prevention, risk factor modification |203204### Output205Generate report per `REPORT_TEMPLATE.md`. See `SCORING_REFERENCE.md` for detailed scoring matrices.206207---208209## Common Use Patterns210211See `EXAMPLES.md` for six detailed worked examples:2121. **Cancer + actionable mutation**: Breast cancer, BRCA1, ER+/HER2- -> Score ~55-65 (HIGH)2132. **Metabolic + PGx concern**: T2D, CYP2C19 PM on clopidogrel -> Score ~55-65 (HIGH)2143. **NSCLC comprehensive**: EGFR L858R, TMB 25, PD-L1 80% -> Score ~75-85 (VERY HIGH)2154. **CVD risk**: LDL 190, SLCO1B1*5, family hx MI -> Score ~50-60 (HIGH)2165. **Rare disease**: Marfan, FBN1 variant -> Score ~55-65 (HIGH)2176. **Neurological risk**: APOE e4/e4, family hx Alzheimer's -> Score ~60-72 (HIGH)