Tooluniverse Immunotherapy Response Prediction

Immunotherapy Response Prediction

Predict patient response to immune checkpoint inhibitors (ICIs) using multi-biomarker integration. Transforms a patient tumor profile (cancer type + mutations + biomarkers) into a quantitative ICI Response Score with drug-specific recommendations, resistance risk assessment, and monitoring plan.

KEY PRINCIPLES:

1. Report-first approach - Create report file FIRST, then populate progressively
2. Evidence-graded - Every finding has an evidence tier (T1-T4)
3. Quantitative output - ICI Response Score (0-100) with transparent component breakdown
4. Cancer-specific - All thresholds and predictions are cancer-type adjusted
5. Multi-biomarker - Integrate TMB + MSI + PD-L1 + neoantigen + mutations
6. Resistance-aware - Always check for known resistance mutations (STK11, PTEN, JAK1/2, B2M)
7. Drug-specific - Recommend specific ICI agents with evidence
8. Source-referenced - Every statement cites the tool/database source
9. English-first queries - Always use English terms in tool calls

When to Use

Apply when user asks:

• "Will this patient respond to immunotherapy?"
• "Should I give pembrolizumab to this melanoma patient?"
• "Patient has NSCLC with TMB 25, PD-L1 80% - predict ICI response"
• "MSI-high colorectal cancer - which checkpoint inhibitor?"
• "Patient has BRAF V600E melanoma, TMB 15 - immunotherapy or targeted?"
• "Low TMB NSCLC with STK11 mutation - should I try immunotherapy?"
• "Compare pembrolizumab vs nivolumab for this patient profile"
• "What biomarkers predict checkpoint inhibitor response?"

Input Parsing

Required: Cancer type + at least one of: mutation list OR TMB value Optional: PD-L1 expression, MSI status, immune infiltration data, HLA type, prior treatments, intended ICI

Accepted Input Formats

Cancer Type Normalization

Common aliases to resolve:

• NSCLC -> non-small cell lung carcinoma
• SCLC -> small cell lung carcinoma
• CRC -> colorectal cancer
• RCC -> renal cell carcinoma
• HNSCC -> head and neck squamous cell carcinoma
• UC / bladder -> urothelial carcinoma
• HCC -> hepatocellular carcinoma
• TNBC -> triple-negative breast cancer
• GEJ -> gastroesophageal junction cancer

Gene Symbol Normalization

• PD-L1 -> CD274
• PD-1 -> PDCD1
• CTLA-4 -> CTLA4
• HER2 -> ERBB2
• MSH2/MLH1/MSH6/PMS2 -> MMR genes

Phase 0: Tool Parameter Reference (CRITICAL)

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

Verified Tool Parameters

Workflow Overview

Input: Cancer type + Mutations/TMB + Optional biomarkers (PD-L1, MSI, etc.)

Phase 1: Input Standardization & Cancer Context
  - Resolve cancer type to EFO ID
  - Parse mutation list
  - Resolve genes to Ensembl/Entrez IDs
  - Get cancer-specific ICI baseline

Phase 2: TMB Analysis
  - TMB classification (low/intermediate/high)
  - Cancer-specific TMB thresholds
  - FDA TMB-H biomarker status

Phase 3: Neoantigen Analysis
  - Estimate neoantigen burden from mutations
  - Mutation type classification (missense/frameshift/nonsense)
  - Neoantigen quality indicators

Phase 4: MSI/MMR Status Assessment
  - MSI status integration
  - MMR gene mutation check
  - FDA MSI-H approval status

Phase 5: PD-L1 Expression Analysis
  - PD-L1 level classification
  - Cancer-specific PD-L1 thresholds
  - FDA-approved PD-L1 cutoffs

Phase 6: Immune Microenvironment Profiling
  - Immune checkpoint gene expression
  - Tumor immune classification (hot/cold)
  - Immune escape signatures

Phase 7: Mutation-Based Predictors
  - Driver mutation analysis
  - Resistance mutations (STK11, PTEN, JAK1/2, B2M)
  - Sensitivity mutations (POLE)
  - DNA damage repair pathway

Phase 8: Clinical Evidence & ICI Options
  - FDA-approved ICIs for this cancer
  - Clinical trial response rates
  - Drug mechanism comparison
  - Combination therapy evidence

Phase 9: Resistance Risk Assessment
  - Known resistance factors
  - Tumor immune evasion mechanisms
  - Prior treatment context

Phase 10: Multi-Biomarker Score Integration
  - Calculate ICI Response Score (0-100)
  - Component breakdown
  - Confidence level

Phase 11: Clinical Recommendations
  - ICI drug recommendation
  - Monitoring plan
  - Alternative strategies

Phase 1: Input Standardization & Cancer Context

Step 1.1: Resolve Cancer Type

# Get cancer EFO ID
result = tu.tools.OpenTargets_get_disease_id_description_by_name(diseaseName='melanoma')
# -> {data: {search: {hits: [{id: 'EFO_0000756', name: 'melanoma', description: '...'}]}}}

Cancer-specific ICI context (hardcoded knowledge base):

Step 1.2: Parse Mutations

Parse each mutation into structured format:

"BRAF V600E" -> {gene: "BRAF", variant: "V600E", type: "missense"}
"TP53 R273H" -> {gene: "TP53", variant: "R273H", type: "missense"}
"STK11 loss" -> {gene: "STK11", variant: "loss of function", type: "loss"}

Step 1.3: Resolve Gene IDs

# For each gene in mutation list
result = tu.tools.MyGene_query_genes(query='BRAF')
# -> hits[0]: {_id: '673', symbol: 'BRAF', ensembl: {gene: 'ENSG00000157764'}}

Phase 2: TMB Analysis

Step 2.1: TMB Classification

If TMB value provided directly, classify:

If only mutations provided, estimate TMB:

• Count total mutations provided
• Note: User-provided lists are typically key mutations, not full exome
• Flag as "estimated from provided mutations - clinical TMB testing recommended"

Step 2.2: TMB FDA Context

# Check FDA TMB-H biomarker approval
result = tu.tools.fda_pharmacogenomic_biomarkers(drug_name='pembrolizumab', limit=100)
# Look for "Tumor Mutational Burden" in Biomarker field
# -> Pembrolizumab approved for TMB-H (>=10 mut/Mb) tissue-agnostic

Step 2.3: Cancer-Specific TMB Thresholds

IMPORTANT: RCC responds to ICIs despite low TMB. TMB is less predictive in some cancers.

Phase 3: Neoantigen Analysis

Step 3.1: Neoantigen Burden Estimation

From mutation list:

• Missense mutations -> Each has ~20-50% chance of generating a neoantigen
• Frameshift mutations -> High neoantigen-generating potential (novel peptides)
• Nonsense mutations -> Moderate potential (truncated proteins)
• Splice site mutations -> Moderate potential (aberrant peptides)

Estimate: neoantigen_count ~= missense_count * 0.3 + frameshift_count * 1.5

Step 3.2: Neoantigen Quality Assessment

# Check mutation impact using UniProt
result = tu.tools.UniProt_get_function_by_accession(accession='P15056')  # BRAF UniProt
# Assess if mutation is in functional domain

Quality indicators:

• Mutations in protein kinase domains -> high immunogenicity potential
• Mutations in surface-exposed regions -> better MHC presentation
• POLE/POLD1 mutations -> ultra-high neoantigen load (ultramutated)

Step 3.3: IEDB Epitope Data (if relevant)

# Check known epitopes for mutated proteins
result = tu.tools.iedb_search_epitopes(organism_name='homo sapiens', source_antigen_name='BRAF')
# Returns known epitopes, MHC restrictions

Neoantigen Score Component

Phase 4: MSI/MMR Status Assessment

Step 4.1: MSI Status Integration

If MSI status provided directly:

Step 4.2: MMR Gene Mutation Check

Check if any provided mutations are in MMR genes:

• MLH1 (ENSG00000076242) - mismatch repair
• MSH2 (ENSG00000095002) - mismatch repair
• MSH6 (ENSG00000116062) - mismatch repair
• PMS2 (ENSG00000122512) - mismatch repair
• EPCAM (ENSG00000119888) - can silence MSH2

If MMR gene mutations found but MSI status not provided -> flag as "possible MSI-H, recommend testing"

Step 4.3: FDA MSI-H Approvals

# Check FDA approvals for MSI-H
result = tu.tools.fda_pharmacogenomic_biomarkers(biomarker='Microsatellite Instability', limit=100)
# Pembrolizumab: tissue-agnostic for MSI-H/dMMR
# Nivolumab: CRC (MSI-H)
# Dostarlimab: dMMR solid tumors

Phase 5: PD-L1 Expression Analysis

Step 5.1: PD-L1 Level Classification

Step 5.2: Cancer-Specific PD-L1 Thresholds

Step 5.3: PD-L1 Gene Expression (Baseline Reference)

# PD-L1 (CD274) expression patterns
result = tu.tools.HPA_get_cancer_prognostics_by_gene(gene_name='CD274')
# Cancer-type specific prognostic data

Phase 6: Immune Microenvironment Profiling

Step 6.1: Key Immune Checkpoint Genes

Query expression data for immune microenvironment markers:

# Key immune genes to check
immune_genes = ['CD274', 'PDCD1', 'CTLA4', 'LAG3', 'HAVCR2', 'TIGIT', 'CD8A', 'CD8B', 'GZMA', 'GZMB', 'PRF1', 'IFNG']

# For each gene, get cancer-specific expression
for gene in immune_genes:
    result = tu.tools.HPA_get_cancer_prognostics_by_gene(gene_name=gene)

Step 6.2: Tumor Immune Classification

Based on available data, classify:

Step 6.3: Immune Pathway Enrichment

# If mutation list includes immune-related genes, do pathway analysis
result = tu.tools.enrichr_gene_enrichment_analysis(
    gene_list=['CD274', 'PDCD1', 'CTLA4', 'IFNG', 'CD8A'],
    libs=['KEGG_2021_Human', 'Reactome_2022']
)

Phase 7: Mutation-Based Predictors

Step 7.1: ICI-Resistance Mutations (CRITICAL)

Known resistance mutations - apply PENALTIES:

Step 7.2: ICI-Sensitivity Mutations (BONUS)

Step 7.3: Driver Mutation Context

# For each mutation, check CIViC evidence for ICI context
# Use OpenTargets for drug associations
result = tu.tools.OpenTargets_get_associated_drugs_by_disease_efoId(efoId='EFO_0000756', size=50)
# Filter for ICI drugs (pembro, nivo, ipi, atezo, durva, avelumab, cemiplimab)

Step 7.4: DNA Damage Repair (DDR) Pathway

Check if mutations are in DDR genes (associated with ICI response):

• ATM, ATR, CHEK1, CHEK2 - DNA damage sensing
• BRCA1, BRCA2, PALB2 - homologous recombination
• RAD50, MRE11, NBN - double-strand break repair
• POLE, POLD1 - polymerase proofreading

DDR mutations -> likely higher TMB -> better ICI response

Phase 8: Clinical Evidence & ICI Options

Step 8.1: FDA-Approved ICIs

# Get FDA indications for key ICIs
ici_drugs = ['pembrolizumab', 'nivolumab', 'atezolizumab', 'durvalumab', 'ipilimumab', 'avelumab', 'cemiplimab']

for drug in ici_drugs:
    result = tu.tools.FDA_get_indications_by_drug_name(drug_name=drug, limit=3)
    # Extract cancer-specific indications

Step 8.2: ICI Drug Profiles

Step 8.3: Clinical Trial Evidence

# Search for ICI trials in this cancer type
result = tu.tools.clinical_trials_search(
    action='search_studies',
    condition='melanoma',
    intervention='pembrolizumab',
    limit=10
)
# Returns: {total_count, studies: [{nctId, title, status, conditions}]}

Step 8.4: Literature Evidence

# Search PubMed for biomarker-specific ICI response data
result = tu.tools.PubMed_search_articles(
    query='pembrolizumab melanoma TMB response biomarker',
    max_results=10
)
# Returns list of {pmid, title, ...}

Step 8.5: OpenTargets Drug-Target Evidence

# Get drug mechanism details
result = tu.tools.OpenTargets_get_drug_mechanisms_of_action_by_chemblId(chemblId='CHEMBL3137343')
# -> pembrolizumab: PD-1 inhibitor, targets PDCD1 (ENSG00000188389)

Key ICI ChEMBL IDs

Phase 9: Resistance Risk Assessment

Step 9.1: Known Resistance Factors Check

For each mutation in the patient profile, check against resistance database:

# Check for resistance evidence in CIViC
# CIViC evidence types: PREDICTIVE, PROGNOSTIC, DIAGNOSTIC, PREDISPOSING, ONCOGENIC
result = tu.tools.civic_search_evidence_items(therapy_name='pembrolizumab')
# Filter for resistance-associated evidence

Step 9.2: Pathway-Level Resistance

Step 9.3: Resistance Risk Score

Summarize resistance risk as:

• Low risk: No resistance mutations, favorable TME
• Moderate risk: 1 resistance factor OR uncertain TME
• High risk: Multiple resistance mutations OR known resistant phenotype

Phase 10: Multi-Biomarker Score Integration

ICI Response Score Calculation (0-100)

TOTAL SCORE = TMB_score + MSI_score + PDL1_score + Neoantigen_score + Mutation_bonus + Resistance_penalty

Where:
  TMB_score:        5-30 points (based on TMB classification)
  MSI_score:        5-25 points (based on MSI status)
  PDL1_score:       5-20 points (based on PD-L1 level)
  Neoantigen_score: 5-15 points (based on estimated neoantigens)
  Mutation_bonus:   0-10 points (POLE, PBRM1, etc.)
  Resistance_penalty: -20 to 0 points (STK11, PTEN, JAK1/2, B2M)

Minimum score: 0 (floor)
Maximum score: 100 (cap)

Response Likelihood Tiers

Confidence Level

Extended Reference: For detailed tool tables, examples, and templates, read REFERENCE.md in this skill directory. The agent can access it via: read skills/tooluniverse-immunotherapy-response-prediction/REFERENCE.md