Precision Oncology Treatment Advisor

Provide actionable treatment recommendations for cancer patients based on their molecular profile using CIViC, ClinVar, OpenTargets, ClinicalTrials.gov, and structure-based analysis.

KEY PRINCIPLES:

1. Report-first - Create report file FIRST, update progressively
2. Evidence-graded - Every recommendation has evidence level
3. Actionable output - Prioritized treatment options, not data dumps
4. Clinical focus - Answer "what should we do?" not "what exists?"
5. English-first queries - Always use English terms in tool calls (mutations, drug names, cancer types), even if the user writes in another language. Only try original-language terms as a fallback. Respond in the user's language

When to Use

Apply when user asks:

• "Patient has [cancer] with [mutation] - what treatments?"
• "What are options for EGFR-mutant lung cancer?"
• "Patient failed [drug], what's next?"
• "Clinical trials for KRAS G12C?"
• "Why isn't [drug] working anymore?"

Phase 0: Tool Verification

CRITICAL: Verify tool parameters before first use.

Workflow Overview

Input: Cancer type + Molecular profile (mutations, fusions, amplifications)

Phase 1: Profile Validation
├── Validate variant nomenclature
├── Resolve gene identifiers
└── Confirm cancer type (EFO/ICD)

Phase 2: Variant Interpretation
├── CIViC → Evidence for each variant
├── ClinVar → Pathogenicity
├── COSMIC → Somatic mutation frequency
├── GDC/TCGA → Real tumor data
├── DepMap → Target essentiality
├── OncoKB → FDA actionability levels (NEW)
├── cBioPortal → Cross-study mutation data (NEW)
├── Human Protein Atlas → Expression validation (NEW)
├── OpenTargets → Target-disease evidence
└── OUTPUT: Variant significance table + target validation + expression

Phase 2.5: Tumor Expression Context (NEW)
├── CELLxGENE → Cell-type specific expression in tumor
├── ChIPAtlas → Regulatory context
├── Cancer-specific expression patterns
└── OUTPUT: Expression validation

Phase 3: Treatment Options
├── Approved therapies (FDA label)
├── NCCN-recommended (literature)
├── Off-label with evidence
└── OUTPUT: Prioritized treatment list

Phase 3.5: Pathway & Network Analysis (NEW)
├── KEGG/Reactome → Pathway context
├── IntAct → Protein interactions
├── Drug combination rationale
└── OUTPUT: Biological context for combinations

Phase 4: Resistance Analysis (if prior therapy)
├── Known resistance mechanisms
├── Structure-based analysis (NvidiaNIM)
├── Network-based bypass pathways (IntAct)
└── OUTPUT: Resistance explanation + strategies

Phase 5: Clinical Trial Matching
├── Active trials for indication + biomarker
├── Eligibility filtering
└── OUTPUT: Matched trials

Phase 5.5: Literature Evidence (NEW)
├── PubMed → Published evidence
├── BioRxiv/MedRxiv → Recent preprints
├── OpenAlex → Citation analysis
└── OUTPUT: Supporting literature

Phase 6: Report Synthesis
├── Executive summary
├── Treatment recommendations (prioritized)
└── Next steps

Phase 1: Profile Validation

1.1 Resolve Gene Identifiers

python
1def resolve_gene(tu, gene_symbol):
2    """Resolve gene to all needed IDs."""
3    ids = {}
4    
5    # Ensembl ID (for OpenTargets)
6    gene_info = tu.tools.MyGene_query_genes(q=gene_symbol, species="human")
7    ids['ensembl'] = gene_info.get('ensembl', {}).get('gene')
8    
9    # UniProt (for structure)
10    uniprot = tu.tools.UniProt_search(query=gene_symbol, organism="human")
11    ids['uniprot'] = uniprot[0].get('primaryAccession') if uniprot else None
12    
13    # ChEMBL target
14    target = tu.tools.ChEMBL_search_targets(query=gene_symbol, organism="Homo sapiens")
15    ids['chembl_target'] = target[0].get('target_chembl_id') if target else None
16    
17    return ids

1.2 Validate Variant Nomenclature

• HGVS protein: p.L858R, p.V600E
• cDNA: c.2573T>G
• Common names: T790M, G12C

Phase 2: Variant Interpretation

2.1 CIViC Evidence Query

python
1def get_civic_evidence(tu, gene_symbol, variant_name):
2    """Get CIViC evidence for variant."""
3    # Search for variant
4    variants = tu.tools.civic_search_variants(query=f"{gene_symbol} {variant_name}")
5    
6    evidence_items = []
7    for var in variants:
8        # Get evidence items for this variant
9        evi = tu.tools.civic_get_variant(id=var['id'])
10        evidence_items.extend(evi.get('evidence_items', []))
11    
12    # Categorize by evidence type
13    return {
14        'predictive': [e for e in evidence_items if e['evidence_type'] == 'Predictive'],
15        'prognostic': [e for e in evidence_items if e['evidence_type'] == 'Prognostic'],
16        'diagnostic': [e for e in evidence_items if e['evidence_type'] == 'Diagnostic']
17    }

2.2 COSMIC Somatic Mutation Analysis (NEW)

python
1def get_cosmic_mutations(tu, gene_symbol, variant_name=None):
2    """Get somatic mutation data from COSMIC database."""
3    
4    # Get all mutations for gene
5    gene_mutations = tu.tools.COSMIC_get_mutations_by_gene(
6        operation="get_by_gene",
7        gene=gene_symbol,
8        max_results=100,
9        genome_build=38
10    )
11    
12    # If specific variant, search for it
13    if variant_name:
14        specific = tu.tools.COSMIC_search_mutations(
15            operation="search",
16            terms=f"{gene_symbol} {variant_name}",
17            max_results=20
18        )
19        return {
20            'specific_variant': specific.get('results', []),
21            'all_gene_mutations': gene_mutations.get('results', [])
22        }
23    
24    return gene_mutations
25
26def get_cosmic_hotspots(tu, gene_symbol):
27    """Identify mutation hotspots in COSMIC."""
28    mutations = tu.tools.COSMIC_get_mutations_by_gene(
29        operation="get_by_gene",
30        gene=gene_symbol,
31        max_results=500
32    )
33    
34    # Count by position
35    position_counts = Counter(m['MutationAA'] for m in mutations.get('results', []))
36    hotspots = position_counts.most_common(10)
37    
38    return hotspots

Why COSMIC matters:

• Gold standard for somatic cancer mutations
• Provides cancer type distribution (which cancers have this mutation)
• FATHMM pathogenicity prediction for novel variants
• Identifies hotspots vs. rare mutations

2.3 GDC/TCGA Pan-Cancer Analysis (NEW)

Access real patient tumor data from The Cancer Genome Atlas:

python
1def get_tcga_mutation_data(tu, gene_symbol, cancer_type=None):
2    """
3    Get somatic mutations from TCGA via GDC.
4    
5    Answers: "How often is this mutation seen in real tumors?"
6    """
7    
8    # Get mutation frequency across all TCGA
9    frequency = tu.tools.GDC_get_mutation_frequency(
10        gene_symbol=gene_symbol
11    )
12    
13    # Get specific mutations
14    mutations = tu.tools.GDC_get_ssm_by_gene(
15        gene_symbol=gene_symbol,
16        project_id=f"TCGA-{cancer_type}" if cancer_type else None,
17        size=50
18    )
19    
20    return {
21        'frequency': frequency.get('data', {}),
22        'mutations': mutations.get('data', {}),
23        'note': 'Real patient tumor data from TCGA'
24    }
25
26def get_tcga_expression_profile(tu, gene_symbol, cancer_type):
27    """Get gene expression data from TCGA."""
28    
29    # Map cancer type to TCGA project
30    project_map = {
31        'lung': 'TCGA-LUAD',
32        'breast': 'TCGA-BRCA', 
33        'colorectal': 'TCGA-COAD',
34        'melanoma': 'TCGA-SKCM',
35        'glioblastoma': 'TCGA-GBM'
36    }
37    project_id = project_map.get(cancer_type.lower(), f'TCGA-{cancer_type.upper()}')
38    
39    expression = tu.tools.GDC_get_gene_expression(
40        project_id=project_id,
41        size=20
42    )
43    
44    return expression.get('data', {})
45
46def get_tcga_cnv_status(tu, gene_symbol, cancer_type):
47    """Get copy number status from TCGA."""
48    
49    project_map = {
50        'lung': 'TCGA-LUAD',
51        'breast': 'TCGA-BRCA'
52    }
53    project_id = project_map.get(cancer_type.lower(), f'TCGA-{cancer_type.upper()}')
54    
55    cnv = tu.tools.GDC_get_cnv_data(
56        project_id=project_id,
57        gene_symbol=gene_symbol,
58        size=20
59    )
60    
61    return cnv.get('data', {})

GDC Tools Summary:

Why TCGA/GDC matters:

• Real patient data - Not cell line or curated, actual tumor sequencing
• Pan-cancer view - Same gene across 33 cancer types
• Multi-omic - Mutations, expression, CNV together
• Clinical correlation - Survival data available

2.4 DepMap Target Validation (NEW)

Assess gene essentiality using CRISPR knockout data from cancer cell lines:

python
1def assess_target_essentiality(tu, gene_symbol, cancer_type=None):
2    """
3    Is this gene essential in cancer cell lines?
4    
5    Essential genes have negative dependency scores.
6    Answers: "If we target this gene, will cancer cells die?"
7    """
8    
9    # Get gene dependency data
10    dependencies = tu.tools.DepMap_get_gene_dependencies(
11        gene_symbol=gene_symbol
12    )
13    
14    # Get cell lines for specific cancer type
15    if cancer_type:
16        cell_lines = tu.tools.DepMap_get_cell_lines(
17            cancer_type=cancer_type,
18            page_size=20
19        )
20        return {
21            'gene': gene_symbol,
22            'dependencies': dependencies.get('data', {}),
23            'cell_lines': cell_lines.get('data', {}),
24            'interpretation': 'Negative scores = gene is essential for cell survival'
25        }
26    
27    return dependencies
28
29def get_depmap_drug_sensitivity(tu, drug_name, cancer_type=None):
30    """Get drug sensitivity data from DepMap."""
31    
32    drugs = tu.tools.DepMap_get_drug_response(
33        drug_name=drug_name
34    )
35    
36    return drugs.get('data', {})

DepMap Tools Summary:

Why DepMap matters for Precision Oncology:

• Target validation - Proves gene is essential for cancer survival
• Cancer selectivity - Essential in cancer but not normal cells?
• Resistance prediction - What other genes become essential when you knockout target?
• Combination rationale - Identify synthetic lethal partners

Example Clinical Application:

markdown
1### Target Essentiality Assessment (DepMap)
2
3**KRAS dependency in pancreatic cancer cell lines**:
4| Cell Line | KRAS Effect Score | Interpretation |
5|-----------|-------------------|----------------|
6| PANC-1 | -0.82 | Strongly essential |
7| MIA PaCa-2 | -0.75 | Essential |
8| BxPC-3 | -0.21 | Less dependent (KRAS WT) |
9
10*Interpretation: KRAS-mutant pancreatic cancer lines are highly dependent on KRAS - validates targeting strategy.*
11
12*Source: DepMap via `DepMap_get_gene_dependencies`*

2.5 OncoKB Actionability Assessment (NEW)

OncoKB provides FDA-approved therapeutic actionability annotations:

python
1def get_oncokb_annotations(tu, gene_symbol, variant_name, tumor_type=None):
2    """
3    Get OncoKB actionability annotations.
4    
5    OncoKB Level of Evidence:
6    - Level 1: FDA-approved
7    - Level 2: Standard care
8    - Level 3A: Compelling clinical evidence
9    - Level 3B: Standard care in different tumor type
10    - Level 4: Biological evidence
11    - R1/R2: Resistance evidence
12    """
13    
14    # Annotate the specific variant
15    annotation = tu.tools.OncoKB_annotate_variant(
16        operation="annotate_variant",
17        gene=gene_symbol,
18        variant=variant_name,  # e.g., "V600E"
19        tumor_type=tumor_type  # OncoTree code e.g., "MEL", "LUAD"
20    )
21    
22    result = {
23        'oncogenic': annotation.get('data', {}).get('oncogenic'),
24        'mutation_effect': annotation.get('data', {}).get('mutationEffect'),
25        'highest_sensitive_level': annotation.get('data', {}).get('highestSensitiveLevel'),
26        'treatments': annotation.get('data', {}).get('treatments', [])
27    }
28    
29    # Get gene-level info
30    gene_info = tu.tools.OncoKB_get_gene_info(
31        operation="get_gene_info",
32        gene=gene_symbol
33    )
34    
35    result['is_oncogene'] = gene_info.get('data', {}).get('oncogene', False)
36    result['is_tumor_suppressor'] = gene_info.get('data', {}).get('tsg', False)
37    
38    return result
39
40def get_oncokb_cnv_annotation(tu, gene_symbol, alteration_type, tumor_type=None):
41    """Get OncoKB annotation for copy number alterations."""
42    
43    annotation = tu.tools.OncoKB_annotate_copy_number(
44        operation="annotate_copy_number",
45        gene=gene_symbol,
46        copy_number_type=alteration_type,  # "AMPLIFICATION" or "DELETION"
47        tumor_type=tumor_type
48    )
49    
50    return {
51        'oncogenic': annotation.get('data', {}).get('oncogenic'),
52        'treatments': annotation.get('data', {}).get('treatments', [])
53    }

OncoKB Level Mapping:

2.6 cBioPortal Cross-Study Analysis (NEW)

Aggregate mutation data across multiple cancer studies:

python
1def get_cbioportal_mutations(tu, gene_symbols, study_id="brca_tcga"):
2    """
3    Get mutation data from cBioPortal across cancer studies.
4    
5    Provides: Mutation types, protein changes, co-mutations.
6    """
7    
8    # Get mutations for genes in study
9    mutations = tu.tools.cBioPortal_get_mutations(
10        study_id=study_id,
11        gene_list=",".join(gene_symbols)  # e.g., "EGFR,KRAS"
12    )
13    
14    # Parse results
15    results = []
16    for mut in mutations or []:
17        results.append({
18            'gene': mut.get('gene', {}).get('hugoGeneSymbol'),
19            'protein_change': mut.get('proteinChange'),
20            'mutation_type': mut.get('mutationType'),
21            'sample_id': mut.get('sampleId'),
22            'validation_status': mut.get('validationStatus')
23        })
24    
25    return results
26
27def get_cbioportal_cancer_studies(tu, cancer_type=None):
28    """Get available cancer studies from cBioPortal."""
29    
30    studies = tu.tools.cBioPortal_get_cancer_studies(limit=50)
31    
32    if cancer_type:
33        studies = [s for s in studies if cancer_type.lower() in s.get('cancerTypeId', '').lower()]
34    
35    return studies
36
37def analyze_co_mutations(tu, gene_symbol, study_id):
38    """Find frequently co-mutated genes."""
39    
40    # Get molecular profiles
41    profiles = tu.tools.cBioPortal_get_molecular_profiles(study_id=study_id)
42    
43    # Get mutation data
44    mutations = tu.tools.cBioPortal_get_mutations(
45        study_id=study_id,
46        gene_list=gene_symbol
47    )
48    
49    return {
50        'profiles': profiles,
51        'mutations': mutations,
52        'study_id': study_id
53    }

cBioPortal Use Cases:

2.7 Human Protein Atlas Expression (NEW)

Validate target expression in tumor vs normal tissues:

python
1def get_hpa_expression(tu, gene_symbol):
2    """
3    Get protein expression data from Human Protein Atlas.
4    
5    Critical for validating:
6    - Target is expressed in tumor tissue
7    - Target has differential tumor vs normal expression
8    """
9    
10    # Search for gene
11    gene_info = tu.tools.HPA_search_genes_by_query(search_query=gene_symbol)
12    
13    if not gene_info:
14        return None
15    
16    # Get tissue expression data
17    ensembl_id = gene_info[0].get('Ensembl') if gene_info else None
18    
19    # Comparative expression in cancer cell lines
20    cell_line_data = tu.tools.HPA_get_comparative_expression_by_gene_and_cellline(
21        gene_name=gene_symbol,
22        cell_line="a549"  # Lung cancer cell line
23    )
24    
25    return {
26        'gene_info': gene_info,
27        'cell_line_expression': cell_line_data
28    }
29
30def check_tumor_specific_expression(tu, gene_symbol, cancer_type):
31    """Check if target has tumor-specific expression pattern."""
32    
33    # Map cancer type to cell line
34    cancer_to_cellline = {
35        'lung': 'a549',
36        'breast': 'mcf7',
37        'liver': 'hepg2',
38        'cervical': 'hela',
39        'prostate': 'pc3'
40    }
41    
42    cell_line = cancer_to_cellline.get(cancer_type.lower(), 'a549')
43    
44    expression = tu.tools.HPA_get_comparative_expression_by_gene_and_cellline(
45        gene_name=gene_symbol,
46        cell_line=cell_line
47    )
48    
49    return expression

HPA Expression Validation Output:

markdown
1### Expression Validation (Human Protein Atlas)
2
3| Gene | Tumor Cell Line | Expression | Normal Tissue | Differential |
4|------|-----------------|------------|---------------|--------------|
5| EGFR | A549 (lung) | High | Low-Medium | Tumor-elevated |
6| ALK | H3122 (lung) | High | Not detected | Tumor-specific |
7| HER2 | MCF7 (breast) | Medium | Low | Elevated |
8
9*Source: Human Protein Atlas via `HPA_get_comparative_expression_by_gene_and_cellline`*

2.8 Evidence Level Mapping

2.4 Output Table

markdown
1## Variant Interpretation
2
3| Variant | Gene | Significance | Evidence Level | Clinical Implication |
4|---------|------|--------------|----------------|---------------------|
5| L858R | EGFR | Oncogenic driver | ★★★ (Level A) | Sensitive to EGFR TKIs |
6| T790M | EGFR | Resistance | ★★★ (Level A) | Resistant to 1st/2nd gen TKIs |
7
8### COSMIC Mutation Frequency
9
10| Gene | Mutation | COSMIC Count | Primary Cancer Types | FATHMM Prediction |
11|------|----------|--------------|---------------------|-------------------|
12| EGFR | L858R | 15,234 | Lung (85%), Colorectal (5%) | Pathogenic |
13| EGFR | T790M | 8,567 | Lung (95%) | Pathogenic |
14| BRAF | V600E | 45,678 | Melanoma (50%), Colorectal (15%) | Pathogenic |
15
16### TCGA/GDC Patient Tumor Data (NEW)
17
18| Gene | TCGA Project | SSM Cases | CNV Amp | CNV Del | % Samples |
19|------|-------------|-----------|---------|---------|-----------|
20| EGFR | TCGA-LUAD | 156 | 89 | 5 | 28% |
21| EGFR | TCGA-GBM | 45 | 312 | 2 | 57% |
22| KRAS | TCGA-PAAD | 134 | 8 | 1 | 92% |
23
24*Source: GDC via `GDC_get_mutation_frequency`, `GDC_get_cnv_data`*
25
26### DepMap Target Essentiality (NEW)
27
28| Gene | Mean Effect (All) | Mean Effect (Cancer Type) | Selectivity | Interpretation |
29|------|-------------------|---------------------------|-------------|----------------|
30| EGFR | -0.15 | -0.45 (lung) | Cancer-selective | Good target |
31| KRAS | -0.82 | -0.91 (pancreatic) | Essential | Hard to target |
32| MYC | -0.95 | -0.93 | Pan-essential | Challenging target |
33
34*Effect score <-0.5 = strongly essential for cell survival*
35*Source: DepMap via `DepMap_get_gene_dependencies`*
36
37*Combined Sources: CIViC, ClinVar, COSMIC, GDC/TCGA, DepMap*

Phase 2.5: Tumor Expression Context (NEW)

2.5.1 Cell-Type Expression in Tumor (CELLxGENE)

python
1def get_tumor_expression_context(tu, gene_symbol, cancer_type):
2    """Get cell-type specific expression in tumor microenvironment."""
3    
4    # Get expression in tumor and normal cells
5    expression = tu.tools.CELLxGENE_get_expression_data(
6        gene=gene_symbol,
7        tissue=cancer_type  # e.g., "lung", "breast"
8    )
9    
10    # Cell metadata for context
11    cell_metadata = tu.tools.CELLxGENE_get_cell_metadata(
12        gene=gene_symbol
13    )
14    
15    # Identify tumor vs normal expression
16    tumor_expression = [c for c in expression if 'tumor' in c.get('cell_type', '').lower()]
17    normal_expression = [c for c in expression if 'normal' in c.get('cell_type', '').lower()]
18    
19    return {
20        'tumor_expression': tumor_expression,
21        'normal_expression': normal_expression,
22        'ratio': calculate_tumor_normal_ratio(tumor_expression, normal_expression)
23    }

Why it matters:

• Confirms target is expressed in tumor cells (not just stroma)
• Identifies potential resistance from tumor heterogeneity
• Supports drug selection based on expression patterns

2.5.2 Output for Report

markdown
1## 2.5 Tumor Expression Context
2
3### Target Expression in Tumor Microenvironment (CELLxGENE)
4
5| Gene | Tumor Cells | Normal Cells | Tumor/Normal Ratio | Interpretation |
6|------|-------------|--------------|-------------------|----------------|
7| EGFR | High (TPM=85) | Medium (TPM=25) | 3.4x | Good target |
8| MET | Medium (TPM=35) | Low (TPM=8) | 4.4x | Potential bypass |
9| AXL | High (TPM=120) | Low (TPM=15) | 8.0x | Resistance marker |
10
11### Cell Type Distribution
12
13- **EGFR-high cells**: Tumor epithelial (85%), CAFs (10%), immune (5%)
14- **MET-high cells**: Tumor epithelial (70%), endothelial (20%), immune (10%)
15
16**Clinical Relevance**: EGFR highly expressed in tumor epithelial cells. AXL overexpression in tumor suggests potential resistance mechanism.
17
18*Source: CELLxGENE Census*

Phase 3: Treatment Options

3.1 Approved Therapies

Query order:

1. OpenTargets_get_associated_drugs_by_target_ensemblId → Approved drugs
2. DailyMed_search_spls → FDA label details
3. ChEMBL_get_drug_mechanisms_of_action_by_chemblId → Mechanism

3.2 Treatment Prioritization

3.3 Output Format

markdown
1## Treatment Recommendations
2
3### First-Line Options
4**1. Osimertinib (Tagrisso)** ★★★
5- FDA-approved for EGFR T790M+ NSCLC
6- Evidence: AURA3 trial (ORR 71%, mPFS 10.1 mo)
7- Source: FDA label, PMID:27959700
8
9### Second-Line Options
10**2. Combination: Osimertinib + [Agent]** ★★☆
11- Evidence: Phase 2 data
12- Source: NCT04487080

Phase 3.5: Pathway & Network Analysis (NEW)

3.5.1 Pathway Context (KEGG/Reactome)

python
1def get_pathway_context(tu, gene_symbols, cancer_type):
2    """Get pathway context for drug combinations and resistance."""
3    
4    pathway_map = {}
5    for gene in gene_symbols:
6        # KEGG pathways
7        kegg_gene = tu.tools.kegg_find_genes(query=f"hsa:{gene}")
8        if kegg_gene:
9            pathways = tu.tools.kegg_get_gene_info(gene_id=kegg_gene[0]['id'])
10            pathway_map[gene] = pathways.get('pathways', [])
11        
12        # Reactome disease score
13        reactome = tu.tools.reactome_disease_target_score(
14            disease=cancer_type,
15            target=gene
16        )
17        pathway_map[f"{gene}_reactome"] = reactome
18    
19    return pathway_map

3.5.2 Protein Interaction Network (IntAct)

python
1def get_resistance_network(tu, drug_target, bypass_candidates):
2    """Find protein interactions that may mediate resistance."""
3    
4    # Get interaction network for drug target
5    network = tu.tools.intact_get_interaction_network(
6        gene=drug_target,
7        depth=2  # Include 2nd degree connections
8    )
9    
10    # Find bypass pathway candidates in network
11    bypass_in_network = [
12        node for node in network['nodes']
13        if node['gene'] in bypass_candidates
14    ]
15    
16    return {
17        'network': network,
18        'bypass_connections': bypass_in_network,
19        'total_interactors': len(network['nodes'])
20    }

3.5.3 Output for Report

markdown
1## 3.5 Pathway & Network Analysis
2
3### Signaling Pathway Context (KEGG)
4
5| Pathway | Genes Involved | Relevance | Drug Targets |
6|---------|---------------|-----------|--------------|
7| EGFR signaling (hsa04012) | EGFR, MET, ERBB3 | Primary pathway | Osimertinib, Capmatinib |
8| PI3K-AKT (hsa04151) | PIK3CA, AKT1 | Downstream | Alpelisib |
9| RAS-MAPK (hsa04010) | KRAS, BRAF, MEK | Bypass potential | Sotorasib, Trametinib |
10
11### Drug Combination Rationale
12
13**Biological basis for combinations**:
14- EGFR inhibition → compensatory MET activation (60% of cases)
15- **Rationale for EGFR + MET inhibition**: Block primary and bypass pathways
16- Network shows direct EGFR-MET interaction (IntAct: MI-score 0.75)
17
18### Protein Interaction Network (IntAct)
19
20| Target | Direct Interactors | Key Partners | Relevance |
21|--------|-------------------|--------------|-----------|
22| EGFR | 156 | MET, ERBB2, ERBB3, GRB2 | Bypass pathways |
23| MET | 89 | EGFR, HGF, GAB1 | Resistance mediator |
24
25*Source: KEGG, Reactome, IntAct*

Phase 4: Resistance Analysis

4.1 Known Mechanisms (Literature + CIViC)

python
1def analyze_resistance(tu, drug_name, gene_symbol):
2    """Find known resistance mechanisms."""
3    # CIViC resistance evidence
4    resistance = tu.tools.civic_search_evidence_items(
5        drug=drug_name,
6        evidence_type="Predictive",
7        clinical_significance="Resistance"
8    )
9    
10    # Literature search
11    papers = tu.tools.PubMed_search_articles(
12        query=f'"{drug_name}" AND "{gene_symbol}" AND resistance',
13        limit=20
14    )
15    
16    return {'civic': resistance, 'literature': papers}

4.2 Structure-Based Analysis (NvidiaNIM)

When mutation affects drug binding:

python
1def model_resistance_mechanism(tu, gene_ids, mutation, drug_smiles):
2    """Model structural impact of resistance mutation."""
3    # Get/predict structure
4    structure = tu.tools.NvidiaNIM_alphafold2(sequence=wild_type_sequence)
5    
6    # Dock drug to wild-type
7    wt_docking = tu.tools.NvidiaNIM_diffdock(
8        protein=structure['structure'],
9        ligand=drug_smiles,
10        num_poses=5
11    )
12    
13    # Compare binding site changes
14    # Report: "T790M introduces bulky methionine, steric clash with erlotinib"

Phase 5: Clinical Trial Matching

5.1 Search Strategy

python
1def find_trials(tu, condition, biomarker, location=None):
2    """Find matching clinical trials."""
3    # Search with biomarker
4    trials = tu.tools.search_clinical_trials(
5        condition=condition,
6        intervention=biomarker,  # e.g., "EGFR"
7        status="Recruiting",
8        pageSize=50
9    )
10    
11    # Get eligibility for top matches
12    nct_ids = [t['nct_id'] for t in trials[:20]]
13    eligibility = tu.tools.get_clinical_trial_eligibility_criteria(nct_ids=nct_ids)
14    
15    return trials, eligibility

5.2 Output Format

markdown
1## Clinical Trial Options
2
3| NCT ID | Phase | Agent | Biomarker Required | Status | Location |
4|--------|-------|-------|-------------------|--------|----------|
5| NCT04487080 | 2 | Amivantamab + lazertinib | EGFR T790M | Recruiting | US, EU |
6| NCT05388669 | 3 | Patritumab deruxtecan | Prior osimertinib | Recruiting | US |
7
8*Source: ClinicalTrials.gov*

Phase 5.5: Literature Evidence (NEW)

5.5.1 Published Literature (PubMed)

python
1def search_treatment_literature(tu, cancer_type, biomarker, drug_name):
2    """Search for treatment evidence in literature."""
3    
4    # Drug + biomarker combination
5    drug_papers = tu.tools.PubMed_search_articles(
6        query=f'"{drug_name}" AND "{biomarker}" AND "{cancer_type}"',
7        limit=20
8    )
9    
10    # Resistance mechanisms
11    resistance_papers = tu.tools.PubMed_search_articles(
12        query=f'"{drug_name}" AND resistance AND mechanism',
13        limit=15
14    )
15    
16    return {
17        'treatment_evidence': drug_papers,
18        'resistance_literature': resistance_papers
19    }

5.5.2 Preprints (BioRxiv/MedRxiv)

python
1def search_preprints(tu, cancer_type, biomarker):
2    """Search preprints for cutting-edge findings."""
3    
4    # BioRxiv cancer research
5    biorxiv = tu.tools.BioRxiv_search_preprints(
6        query=f"{cancer_type} {biomarker} treatment",
7        limit=10
8    )
9    
10    # MedRxiv clinical studies
11    medrxiv = tu.tools.MedRxiv_search_preprints(
12        query=f"{cancer_type} {biomarker}",
13        limit=10
14    )
15    
16    return {
17        'biorxiv': biorxiv,
18        'medrxiv': medrxiv
19    }

5.5.3 Citation Analysis (OpenAlex)

python
1def analyze_key_papers(tu, key_papers):
2    """Get citation metrics for key evidence papers."""
3    
4    analyzed = []
5    for paper in key_papers[:10]:
6        work = tu.tools.openalex_search_works(
7            query=paper['title'],
8            limit=1
9        )
10        if work:
11            analyzed.append({
12                'title': paper['title'],
13                'citations': work[0].get('cited_by_count', 0),
14                'year': work[0].get('publication_year'),
15                'open_access': work[0].get('is_oa', False)
16            })
17    
18    return analyzed

5.5.4 Output for Report

markdown
1## 5.5 Literature Evidence
2
3### Key Clinical Studies
4
5| PMID | Title | Year | Citations | Evidence Type |
6|------|-------|------|-----------|---------------|
7| 27959700 | AURA3: Osimertinib vs chemotherapy... | 2017 | 2,450 | Phase 3 trial |
8| 30867819 | Mechanisms of osimertinib resistance... | 2019 | 680 | Review |
9| 34125020 | Amivantamab + lazertinib Phase 1... | 2021 | 320 | Phase 1 trial |
10
11### Recent Preprints (Not Peer-Reviewed)
12
13| Source | Title | Posted | Key Finding |
14|--------|-------|--------|-------------|
15| MedRxiv | Novel C797S resistance strategy... | 2024-01 | Fourth-gen TKI |
16| BioRxiv | scRNA-seq reveals resistance... | 2024-02 | Cell state switch |
17
18**⚠️ Note**: Preprints have NOT undergone peer review. Interpret with caution.
19
20### Evidence Summary
21
22| Category | Papers Found | High-Impact (>100 citations) |
23|----------|--------------|------------------------------|
24| Treatment efficacy | 25 | 8 |
25| Resistance mechanisms | 18 | 5 |
26| Combinations | 12 | 3 |
27
28*Source: PubMed, BioRxiv, MedRxiv, OpenAlex*

Report Template

File: [PATIENT_ID]_oncology_report.md

markdown
1# Precision Oncology Report
2
3**Patient ID**: [ID] | **Date**: [Date]
4
5## Patient Profile
6- **Diagnosis**: [Cancer type, stage]
7- **Molecular Profile**: [Mutations, fusions]
8- **Prior Therapy**: [Previous treatments]
9
10---
11
12## Executive Summary
13[2-3 sentence summary of key findings and recommendation]
14
15---
16
17## 1. Variant Interpretation
18[Table with variants, significance, evidence levels]
19
20## 2. Treatment Recommendations
21### First-Line Options
22[Prioritized list with evidence]
23
24### Second-Line Options
25[Alternative approaches]
26
27## 3. Resistance Analysis (if applicable)
28[Mechanism explanation, strategies to overcome]
29
30## 4. Clinical Trial Options
31[Matched trials with eligibility]
32
33## 5. Next Steps
341. [Specific actionable recommendation]
352. [Follow-up testing if needed]
363. [Referral if appropriate]
37
38---
39
40## Data Sources
41| Source | Query | Data Retrieved |
42|--------|-------|----------------|
43| CIViC | [gene] [variant] | Evidence items |
44| ClinicalTrials.gov | [condition] | Active trials |

Completeness Checklist

Before finalizing report:

• All variants interpreted with evidence levels
• ≥1 first-line recommendation with ★★★ evidence (or explain why none)
• Resistance mechanism addressed (if prior therapy failed)
• ≥3 clinical trials listed (or "no matching trials")
• Executive summary is actionable (says what to DO)
• All recommendations have source citations

Fallback Chains

Evidence Grading

Tool Reference

See TOOLS_REFERENCE.md for complete tool documentation.