Tooluniverse Precision Medicine Stratification

Precision Medicine Patient Stratification

Transform patient genomic and clinical profiles into actionable risk stratification, treatment recommendations, and personalized therapeutic strategies. Integrates germline genetics, somatic alterations, pharmacogenomics, pathway biology, and clinical evidence to produce a quantitative risk score with tiered management recommendations.

KEY PRINCIPLES:

1. Report-first approach - Create report file FIRST, then populate progressively
2. Disease-specific logic - Cancer vs metabolic vs rare disease pipelines diverge at Phase 2
3. Multi-level integration - Germline + somatic + expression + clinical data layers
4. Evidence-graded - Every finding has an evidence tier (T1-T4)
5. Quantitative output - Precision Medicine Risk Score (0-100) with transparent components
6. Pharmacogenomic guidance - Drug selection AND dosing recommendations
7. Guideline-concordant - Reference NCCN, ACC/AHA, ADA, and other guidelines
8. Source-referenced - Every statement cites the tool/database source
9. Completeness checklist - Mandatory section showing data availability and analysis coverage
10. English-first queries - Always use English terms in tool calls. Respond in user's language

When to Use

Apply when user asks:

• "Stratify this breast cancer patient: ER+/HER2-, BRCA1 mutation, stage II"
• "What is the risk profile for this diabetes patient with HbA1c 8.5 and CYP2C19 poor metabolizer?"
• "NSCLC patient with EGFR L858R, stage IV, TMB 25 - treatment strategy?"
• "Predict prognosis and recommend treatment for this cardiovascular patient"
• "Patient has Marfan syndrome with FBN1 mutation - risk stratification"
• "Alzheimer's risk assessment: APOE e4/e4, family history positive"
• "Personalized treatment plan for type 2 diabetes with genetic risk factors"
• "Which therapy is best for this patient's molecular profile?"

NOT for (use other skills instead):

• Single variant interpretation -> Use tooluniverse-variant-interpretation or tooluniverse-cancer-variant-interpretation
• Immunotherapy-specific prediction -> Use tooluniverse-immunotherapy-response-prediction
• Drug safety profiling only -> Use tooluniverse-adverse-event-detection
• Target validation -> Use tooluniverse-drug-target-validation
• Clinical trial search only -> Use tooluniverse-clinical-trial-matching
• Drug-drug interaction analysis only -> Use tooluniverse-drug-drug-interaction
• PRS calculation only -> Use tooluniverse-polygenic-risk-score

Input Parsing

Required Input

• Disease/condition: Free-text disease name (e.g., "breast cancer", "type 2 diabetes", "Marfan syndrome")
• At least one of: Germline variants, somatic mutations, gene list, or clinical biomarkers

Strongly Recommended

• Genomic data: Specific variants (e.g., "BRCA1 c.68_69delAG", "EGFR L858R"), gene names, or expression changes
• Clinical parameters: Age, sex, disease stage, biomarkers (HbA1c, PSA, LDL-C)

Optional (improves stratification)

• Comorbidities: Other conditions (e.g., "hypertension", "diabetes")
• Prior treatments: Previous therapies and responses
• Family history: Affected relatives, inheritance pattern
• Ethnicity: For population-specific risk calibration
• Current medications: For DDI and pharmacogenomic analysis
• Stratification goal: Risk assessment, treatment selection, prognosis, prevention

Input Format Examples

Disease Type Classification

Classify the disease into one of these categories (determines Phase 2 routing):

Gene Symbol Normalization

Phase 0: Tool Parameter Reference (CRITICAL)

BEFORE calling ANY tool, verify parameters using this reference table.

Verified Tool Parameters

Response Format Notes

• OpenTargets: Always nested {data: {entity: {field: ...}}} structure
• FDA label tools: Return {meta: {disclaimer, terms, license, ...}, results: [...]}. Access via result['results'][0]['field']
• DrugBank: ALL tools require 4 params: query, case_sensitive (bool), exact_match (bool), limit (int)
• PharmGKB: Returns complex nested objects. Check for data wrapper
• PubMed_search_articles: Returns a plain list of dicts, NOT {articles: [...]}
• ClinVar: clinvar_search_variants returns list of variants with clinical significance
• gnomAD: May return "Service overloaded" - treat as transient, retry or skip
• fda_pharmacogenomic_biomarkers: Default limit=10, use limit=1000 to get all
• cBioPortal_get_mutations: gene_list is a STRING, not array. cBioPortal tools may have URL bugs
• ClinVar: May return either a plain list or {status, data: {esearchresult: {count, idlist}}} - handle both
• EnsemblVEP: May return either a list [{...}] or {data: {...}, metadata: {...}} - handle both
• PubMed_Guidelines_Search: Requires limit parameter (NOT max_results), may require API key. Use PubMed_search_articles as fallback
• gwas_get_associations_for_trait: May return errors; use gwas_search_associations instead
• MyGene CYP2D6: First result may be LOC110740340; always filter by symbol match

Workflow Overview

Input: Disease + Genomic data + Clinical parameters + Stratification goal

Phase 1: Disease Disambiguation & Profile Standardization
  - Resolve disease to EFO/MONDO IDs
  - Classify disease type (cancer/metabolic/CVD/neuro/rare/autoimmune)
  - Parse genomic data (variants, genes, expression)
  - Resolve gene IDs (Ensembl, Entrez, UniProt)

Phase 2: Genetic Risk Assessment
  - Germline variant pathogenicity (ClinVar, VEP)
  - Gene-disease association strength (OpenTargets)
  - GWAS-based polygenic risk estimation
  - Population frequency (gnomAD)
  - Gene constraint/intolerance (gnomAD)

Phase 3: Disease-Specific Molecular Stratification
  CANCER PATH:
    - Molecular subtyping (driver mutations, receptor status)
    - Prognostic markers (stage + grade + molecular)
    - TMB/MSI/HRD assessment
    - Somatic mutation landscape (cBioPortal)
  METABOLIC PATH:
    - Genetic risk + clinical risk integration
    - Complication risk (nephropathy, neuropathy, CVD)
    - Monogenic subtypes (MODY, lipodystrophy)
  CVD PATH:
    - ASCVD risk integration
    - Familial hypercholesterolemia genes
    - Statin/anticoagulant PGx
  RARE DISEASE PATH:
    - Causal variant identification
    - Genotype-phenotype correlation
    - Penetrance estimation

Phase 4: Pharmacogenomic Profiling
  - Drug-metabolizing enzyme genotypes (CYP2D6, CYP2C19, CYP3A4)
  - Drug transporter variants (SLCO1B1, ABCB1)
  - Drug target variants (VKORC1, DPYD, UGT1A1)
  - HLA alleles (drug hypersensitivity risk)
  - PharmGKB clinical annotations
  - FDA pharmacogenomic biomarkers

Phase 5: Comorbidity & Drug Interaction Risk
  - Disease-disease genetic overlap
  - Impact on treatment selection
  - Drug-drug interaction risk
  - Pharmacogenomic DDI amplification

Phase 6: Molecular Pathway Analysis
  - Dysregulated pathway identification (Reactome, KEGG)
  - Network disruption analysis (STRING)
  - Druggable pathway targets
  - Pathway-based therapeutic opportunities

Phase 7: Clinical Evidence & Guidelines
  - Guideline-based risk categories (NCCN, ACC/AHA, ADA)
  - FDA-approved therapies for patient profile
  - Literature evidence (PubMed)
  - Biomarker-guided treatment evidence

Phase 8: Clinical Trial Matching
  - Trials matching molecular profile
  - Biomarker-driven trials
  - Precision medicine basket/umbrella trials
  - Risk-adapted trials

Phase 9: Integrated Scoring & Recommendations
  - Calculate Precision Medicine Risk Score (0-100)
  - Risk tier assignment (Low/Int/High/Very High)
  - Treatment algorithm (1st/2nd/3rd line)
  - Monitoring plan
  - Outcome predictions

Phase 1: Disease Disambiguation & Profile Standardization

Step 1.1: Resolve Disease to EFO ID

# Get disease EFO ID
result = tu.tools.OpenTargets_get_disease_id_description_by_name(diseaseName='breast cancer')
# -> {data: {search: {hits: [{id: 'EFO_0000305', name: 'breast carcinoma', description: '...'}]}}}
efo_id = result['data']['search']['hits'][0]['id']

Common Disease EFO IDs (for reference):

Step 1.2: Classify Disease Type

Based on disease name and EFO ID, classify into: CANCER, METABOLIC, CVD, NEUROLOGICAL, RARE, AUTOIMMUNE. This determines the Phase 3 routing.

Step 1.3: Parse Genomic Data

Parse each variant/gene into structured format:

"BRCA1 c.68_69delAG" -> {gene: "BRCA1", variant: "c.68_69delAG", type: "frameshift"}
"EGFR L858R" -> {gene: "EGFR", variant: "L858R", type: "missense"}
"CYP2C19 *2/*2" -> {gene: "CYP2C19", genotype: "*2/*2", metabolizer_status: "poor"}
"APOE e4/e4" -> {gene: "APOE", genotype: "e4/e4", risk_allele: "e4"}

Step 1.4: Resolve Gene IDs

# For each gene in profile
result = tu.tools.MyGene_query_genes(query='BRCA1')
# -> hits[0]: {_id: '672', symbol: 'BRCA1', ensembl: {gene: 'ENSG00000012048'}}
ensembl_id = result['hits'][0]['ensembl']['gene']
entrez_id = result['hits'][0]['_id']

Critical Gene IDs (pre-resolved):

Phase 2: Genetic Risk Assessment

Step 2.1: Germline Variant Pathogenicity

For each germline variant provided:

# Search ClinVar for variant pathogenicity
result = tu.tools.clinvar_search_variants(gene='BRCA1', significance='pathogenic', limit=50)
# Check if patient's specific variant is in ClinVar

# For rsID variants, get VEP annotation
result = tu.tools.EnsemblVEP_annotate_rsid(variant_id='rs80357906')
# Returns SIFT, PolyPhen predictions, consequence type

# For HGVS variants
result = tu.tools.EnsemblVEP_annotate_hgvs(hgvs_notation='ENST00000357654.9:c.5266dupC', species='homo_sapiens')

Pathogenicity Classification (ACMG-aligned):

Step 2.2: Gene-Disease Association Strength

# Get genetic evidence for gene-disease pair
result = tu.tools.OpenTargets_target_disease_evidence(
    ensemblId='ENSG00000012048',  # BRCA1
    efoId='EFO_0000305',         # breast cancer
    size=20
)
# Returns evidence items with scores

Step 2.3: GWAS-Based Polygenic Risk

# Search GWAS associations for disease
result = tu.tools.gwas_get_associations_for_trait(trait='breast cancer')
# Returns associated SNPs with effect sizes

# Search GWAS studies via OpenTargets
result = tu.tools.OpenTargets_search_gwas_studies_by_disease(
    diseaseIds=['EFO_0000305'], size=25
)

# For specific genes, check GWAS hits
result = tu.tools.GWAS_search_associations_by_gene(gene_name='BRCA1')

PRS Estimation (from available GWAS data):

Note: With user-provided variants only (not full genotype), estimate approximate PRS by counting known risk alleles and their effect sizes from GWAS catalog. Flag as "estimated - full genotyping recommended for precise PRS."

Step 2.4: Population Frequency

# Check variant frequency in gnomAD
result = tu.tools.gnomad_get_variant(variant_id='1-55505647-G-T')
# Returns allele frequency across populations

Step 2.5: Gene Constraint

# Gene intolerance to loss of function
result = tu.tools.gnomad_get_gene_constraints(gene_symbol='BRCA1')
# Returns pLI, LOEUF scores - high pLI/low LOEUF = haploinsufficiency

Genetic Risk Score Component (0-35 points):

Combine pathogenicity + gene-disease association + PRS:

• Pathogenic variant in disease gene: 25+ points
• Strong GWAS associations (multiple risk alleles): up to 35 points
• VUS in relevant gene: 10-15 points
• No known pathogenic variants but some risk alleles: 5-15 points

Phase 3: Disease-Specific Molecular Stratification

CANCER PATH (Phase 3C)

Step 3C.1: Molecular Subtyping

# Get somatic mutation landscape from cBioPortal
result = tu.tools.cBioPortal_get_mutations(
    study_id='brca_tcga_pub',  # breast cancer TCGA
    gene_list='BRCA1 BRCA2 TP53 PIK3CA ESR1 ERBB2'  # STRING, not array
)
# Returns mutation frequencies, types

# Check cancer prognostic markers
result = tu.tools.HPA_get_cancer_prognostics_by_gene(gene_name='ESR1')
# Returns prognostic data for breast cancer

Cancer-Specific Subtype Definitions:

Step 3C.2: TMB/MSI/HRD Assessment

If TMB provided:

# Check FDA TMB-H approvals
result = tu.tools.fda_pharmacogenomic_biomarkers(drug_name='pembrolizumab', limit=100)
# Look for "Tumor Mutational Burden" in Biomarker field

Step 3C.3: Prognostic Stratification

Combine stage + molecular features:

METABOLIC PATH (Phase 3M)

Step 3M.1: Clinical Risk Integration

# Check genetic risk factors for T2D
result = tu.tools.GWAS_search_associations_by_gene(gene_name='TCF7L2')
# TCF7L2 is strongest T2D risk gene

# Check monogenic diabetes genes
result = tu.tools.OpenTargets_target_disease_evidence(
    ensemblId='ENSG00000148737',  # TCF7L2
    efoId='EFO_0001360',         # T2D
    size=20
)

T2D Stratification:

CVD PATH (Phase 3V)

# Check PCSK9 and LDLR variants
result = tu.tools.clinvar_search_variants(gene='LDLR', significance='pathogenic', limit=20)
# Familial hypercholesterolemia check

# Check statin-relevant PGx
result = tu.tools.PharmGKB_get_clinical_annotations(query='SLCO1B1')
# SLCO1B1 *5 -> increased statin myopathy risk

CVD Risk Integration:

RARE DISEASE PATH (Phase 3R)

# Check causal variant in disease gene
result = tu.tools.clinvar_search_variants(gene='FBN1', significance='pathogenic', limit=50)
# Marfan syndrome - FBN1 pathogenic variants

# Genotype-phenotype correlation
result = tu.tools.UniProt_get_disease_variants_by_accession(accession='P35555')  # FBN1 UniProt
# Known disease variants and their phenotypes

Rare Disease Risk Assessment:

Phase 4: Pharmacogenomic Profiling

Step 4.1: Drug-Metabolizing Enzyme Genotypes

# PharmGKB clinical annotations for CYP2C19
result = tu.tools.PharmGKB_get_clinical_annotations(query='CYP2C19')
# Returns drug-gene pairs with clinical annotation levels

# FDA pharmacogenomic biomarkers
result = tu.tools.fda_pharmacogenomic_biomarkers(drug_name='clopidogrel', limit=50)
# CYP2C19 poor metabolizer -> reduced clopidogrel efficacy

# PharmGKB dosing guidelines
result = tu.tools.PharmGKB_get_dosing_guidelines(query='CYP2C19')
# CPIC dosing guidelines

Key Pharmacogenes and Clinical Impact:

Step 4.2: Treatment-Specific PGx

# For the specific disease, identify relevant drugs and check PGx
# Example: breast cancer -> tamoxifen -> CYP2D6
result = tu.tools.PharmGKB_get_drug_details(query='tamoxifen')
# Returns PGx annotations for tamoxifen

# Get FDA PGx biomarkers for disease area
result = tu.tools.fda_pharmacogenomic_biomarkers(biomarker='CYP2D6', limit=100)
# All drugs with CYP2D6 PGx in FDA labels

Step 4.3: Drug Target Variants

# Check if patient has variants in drug targets
result = tu.tools.PharmGKB_search_variants(query='VKORC1')
# VKORC1 variants affecting warfarin response

Pharmacogenomic Risk Score (0-10 points):

• Poor metabolizer for treatment-relevant CYP: 8-10 points
• Intermediate metabolizer: 4-5 points
• High-risk HLA allele: 8-10 points
• Drug target variant: 3-5 points
• Normal metabolizer, no actionable PGx: 0 points

Phase 5: Comorbidity & Drug Interaction Risk

Step 5.1: Comorbidity Analysis

# Check disease-disease overlap via shared genetic targets
result = tu.tools.OpenTargets_get_associated_targets_by_disease_efoId(
    efoId='EFO_0001360',  # T2D
    size=50
)
# Compare top targets between primary disease and comorbidities

# Literature on comorbidity
result = tu.tools.PubMed_search_articles(
    query='type 2 diabetes cardiovascular comorbidity risk',
    max_results=5
)

Step 5.2: Drug-Drug Interaction Risk

# If current medications provided, check DDI
result = tu.tools.drugbank_get_drug_interactions_by_drug_name_or_id(
    query='metformin',
    case_sensitive=False,
    exact_match=False,
    limit=20
)

# FDA DDI data
result = tu.tools.FDA_get_drug_interactions_by_drug_name(drug_name='metformin', limit=5)

Step 5.3: PGx-Amplified DDI Risk

If patient is a CYP2D6 poor metabolizer AND taking a CYP2D6 inhibitor -> compounded risk.

Phase 6: Molecular Pathway Analysis

Step 6.1: Dysregulated Pathways

# Pathway enrichment for affected genes
gene_list = ['BRCA1', 'TP53', 'PIK3CA']  # from patient mutations
result = tu.tools.enrichr_gene_enrichment_analysis(
    gene_list=gene_list,
    libs=['KEGG_2021_Human', 'Reactome_2022']
)
# Returns enriched pathways with p-values

# Reactome pathway analysis
# First get UniProt IDs, then map to pathways
result = tu.tools.Reactome_map_uniprot_to_pathways(id='P38398')  # BRCA1 UniProt
# Returns list of pathways involving BRCA1

Step 6.2: Network Analysis

# Protein-protein interaction network
result = tu.tools.STRING_get_interaction_partners(
    protein_ids=['BRCA1', 'TP53'],
    species=9606,
    limit=20
)

# Functional enrichment of network
result = tu.tools.STRING_functional_enrichment(
    protein_ids=['BRCA1', 'TP53', 'PALB2', 'RAD51'],
    species=9606
)

Step 6.3: Druggable Pathway Targets

# Check tractability of pathway nodes
for gene in pathway_genes:
    result = tu.tools.OpenTargets_get_target_tractability_by_ensemblID(ensemblId=ensembl_id)
    # Returns small molecule, antibody, PROTAC tractability

Key Druggable Pathways:

Phase 7: Clinical Evidence & Guidelines

Step 7.1: Guideline-Based Risk Categories

# Search clinical guidelines in PubMed
result = tu.tools.PubMed_Guidelines_Search(
    query='NCCN breast cancer BRCA1 treatment guidelines',
    max_results=5
)

# Search general evidence
result = tu.tools.PubMed_search_articles(
    query='BRCA1 breast cancer treatment stratification',
    max_results=10
)

Guideline References by Disease:

Step 7.2: FDA-Approved Therapies

# Get approved drugs for disease
result = tu.tools.OpenTargets_get_associated_drugs_by_disease_efoId(
    efoId='EFO_0000305',  # breast cancer
    size=50
)
# Returns all known drugs with clinical status

# Check specific drug FDA info
result = tu.tools.FDA_get_indications_by_drug_name(drug_name='olaparib', limit=5)
# PARP inhibitor for BRCA-mutated breast cancer

# Get drug mechanism
result = tu.tools.FDA_get_mechanism_of_action_by_drug_name(drug_name='olaparib', limit=5)

Step 7.3: Biomarker-Drug Evidence

# CIViC evidence for biomarker-drug pair
result = tu.tools.civic_search_evidence_items(
    therapy_name='olaparib',
    disease_name='breast cancer'
)
# Returns clinical evidence items with evidence levels

# DrugBank for drug details
result = tu.tools.drugbank_get_drug_basic_info_by_drug_name_or_id(
    query='olaparib',
    case_sensitive=False,
    exact_match=False,
    limit=5
)

Phase 8: Clinical Trial Matching

Step 8.1: Biomarker-Driven Trials

# Search trials matching molecular profile
result = tu.tools.clinical_trials_search(
    action='search_studies',
    condition='breast cancer',
    intervention='PARP inhibitor',
    limit=10
)
# Returns {total_count, studies: [{nctId, title, status, conditions}]}

# Alternative search
result = tu.tools.search_clinical_trials(
    query_term='BRCA1 breast cancer',
    condition='breast cancer',
    intervention='olaparib',
    pageSize=10
)

Step 8.2: Precision Medicine Trials

# Search basket/umbrella trials
result = tu.tools.search_clinical_trials(
    query_term='precision medicine biomarker-driven',
    condition='breast cancer',
    pageSize=10
)

# Search risk-adapted trials
result = tu.tools.search_clinical_trials(
    query_term='high risk BRCA1',
    condition='breast cancer',
    pageSize=10
)

Step 8.3: Trial Details

# Get details for promising trials
result = tu.tools.clinical_trials_get_details(
    action='get_study_details',
    nct_id='NCT03344965'
)
# Returns full study protocol

Phase 9: Integrated Scoring & Recommendations

Precision Medicine Risk Score (0-100)

Score Components

Genetic Risk Component (0-35 points):

Clinical Risk Component (0-30 points):

Molecular Features Component (0-25 points):

Pharmacogenomic Risk Component (0-10 points):

Risk Tier Assignment

Treatment Algorithm

Based on disease type + risk tier + molecular profile + PGx:

Cancer Treatment Algorithm

IF actionable mutation present:
  1st line: Targeted therapy (e.g., EGFR TKI, BRAF inhibitor, PARP inhibitor)
  2nd line: Immunotherapy (if TMB-H or MSI-H) OR chemotherapy
  3rd line: Clinical trial OR alternative targeted therapy

IF no actionable mutation:
  IF TMB-H or MSI-H:
    1st line: Immunotherapy (pembrolizumab)
    2nd line: Chemotherapy
  ELSE:
    1st line: Standard chemotherapy (disease-specific)
    2nd line: Consider clinical trials

PGx adjustments:
  - DPYD deficient -> AVOID fluoropyrimidines or reduce d

…(truncated)