Skip to content

Latest commit

 

History

History
522 lines (397 loc) · 17.2 KB

File metadata and controls

522 lines (397 loc) · 17.2 KB

Complete Workflow Prompts

End-to-end multi-modal precision medicine analyses.

Servers used: All MCP servers


PatientOne Complete Analysis

1. Comprehensive Multi-Modal Analysis (25-35 min DRY_RUN / 2-5 hrs production)

Perform comprehensive multi-modal analysis for PatientOne (PAT001-OVC-2025)
to identify top 3 treatment targets:

STEP 1: Clinical Context (5 minutes)
- Get patient demographics from mcp-mockepic
- Extract diagnosis: Stage, histology, grade
- Review treatment history: Prior therapies and responses
- Key biomarkers: CA-125, BRCA status

STEP 2: Genomic Analysis (7 minutes)
- Load VCF file: /data/patient-data/PAT001-OVC-2025/genomic/variants.vcf
- Identify pathogenic variants in cancer genes: TP53, BRCA1, PIK3CA, PTEN
- Interpret clinical significance of each variant
- Map variants to pathways

STEP 3: Multi-Omics Integration (10 minutes)
- Load RNA, protein, phospho data from multiomics folder
- Run Stouffer meta-analysis for concordant changes
- Identify activated pathways (KEGG, Hallmark) with FDR < 0.05
- Find upstream regulators (kinases, transcription factors)

STEP 4: Spatial Transcriptomics (10 minutes)
- Load Visium data from spatial folder
- Perform spatial differential expression (tumor vs. normal)
- Run spatial pathway enrichment focusing on tumor regions
- Identify spatially variable genes (Moran's I, p < 0.05)
- Cell type deconvolution (tumor, stroma, immune)

STEP 5: Treatment Synthesis (3 minutes)
- Map activated pathways to FDA-approved drugs
- Prioritize by: Genomic evidence + Multi-omics + Spatial concordance
- Rank top 3 treatment options with rationale

RETURN: Structured report with:
- Clinical summary (1 paragraph)
- Key genomic findings (bullet list)
- Top 3 activated pathways (with p-values)
- Spatial heterogeneity summary
- Treatment recommendations (ranked 1-3 with evidence levels)
- Visualizations: Volcano plot, pathway heatmap, spatial map

Extend with external sources: After identifying treatment targets, use ClinicalTrials.gov and PubMed connectors to search for active trials and supporting literature. See PROMPT 5b for a detailed example, or Connect External MCP for setup.

Expected Results:

  • Clinical: 58F, Stage IV HGSOC, platinum-resistant
  • Genomic: TP53 mutated, BRCA1 germline variant
  • Pathways: PI3K/AKT/mTOR (p=8.2e-5), DNA repair (p=1.3e-4)
  • Spatial: Hypoxic core, immune exclusion
  • Recommendations:
    1. Olaparib (PARP inhibitor) - BRCA1 variant
    2. Everolimus (mTOR inhibitor) - Pathway activation
    3. Pembrolizumab (Checkpoint inhibitor) - Immune edges

Time: 25-35 min (DRY_RUN) / 2-5 hours (production) Cost: ~$1 (DRY_RUN) / see Cost Analysis for production costs


Research Workflows

2. Tumor Microenvironment Characterization

Characterize the tumor microenvironment for PatientOne:

STEP 1: Spatial Analysis
- Load Visium spatial data
- Identify tumor regions (T1 core, T2 edge) vs. normal (N1)
- Cell type deconvolution: Tumor, stroma, immune, hypoxic
- Spatial neighborhoods: Cluster spots by expression similarity

STEP 2: Cell-Cell Interactions
- Ligand-receptor analysis (tumor-stroma, tumor-immune)
- Identify communication pathways (VEGF, PD-L1/PD-1, TGF-β)
- Map interaction zones spatially

STEP 3: Immune Profiling
- Quantify immune infiltration (T-cells, macrophages)
- Identify immune-excluded vs. inflamed regions
- Check immune checkpoint expression (PD-L1, CTLA4, LAG3)

STEP 4: Imaging Integration
- Load H&E histology slide
- Correlate gene expression with histological features
- Validate cell type annotations with pathologist

RETURN:
- Microenvironment map (cell types, neighborhoods, interactions)
- Immune landscape summary (hot vs. cold vs. excluded)
- Therapeutic implications (immunotherapy candidates, combination strategies)

Use Case: Understanding treatment resistance, immunotherapy candidacy


3. Drug Resistance Mechanism Identification

Identify mechanisms of platinum resistance in PatientOne:

STEP 1: Genomic Resistance Markers
- Check for BRCA1/2 reversion mutations (restore HR)
- Drug efflux pumps: MDR1, MRP1 overexpression
- Cell cycle alterations: TP53, RB1 status

STEP 2: Multi-Omics Resistance Signatures
- Compare tumor pre-treatment vs. progression (if longitudinal)
- Identify upregulated resistance pathways:
  * Drug efflux
  * DNA repair restoration
  * EMT activation
  * Anti-apoptotic signaling

STEP 3: Spatial Heterogeneity
- Map resistance signatures spatially
- Identify resistant subclones (regional expression patterns)
- Check if resistance is uniform or focal

STEP 4: Treatment Bypass Strategies
- Activated pathways not targeted by platinum
- PI3K/AKT/mTOR, MAPK as escape routes
- Suggest combination therapies

RETURN:
- Resistance mechanisms identified (ranked by evidence)
- Spatial map of resistant regions
- Alternative treatment recommendations

Extend with external sources: Search PubMed for resistance mechanism publications and bioRxiv for preprints on emerging bypass strategies. Search ClinicalTrials.gov for trials targeting the identified resistance pathways. See Connect External MCP.

Use Case: Second-line treatment selection, clinical trial matching


4. Biomarker Discovery Workflow

Discover prognostic/predictive biomarkers from PatientOne (as pilot):

STEP 1: Candidate Identification
- Multi-omics differential expression (responders vs. non-responders)
- Spatial biomarkers (immune infiltration, pathway scores)
- Genomic biomarkers (variants, mutational signatures)

STEP 2: Feature Selection
- Rank candidates by:
  * Effect size (fold change, odds ratio)
  * Statistical significance (FDR < 0.01)
  * Biological plausibility (known cancer roles)

STEP 3: Validation Planning
- Select top 10 candidates for validation
- Recommend validation cohort (TCGA ovarian cancer)
  TIP: Use cBioPortal connector to query TCGA-OV cohort data directly.
  See [Connect External MCP](../../for-researchers/CONNECT_EXTERNAL_MCP.md).
- Suggest assay platforms (IHC for protein, qPCR for RNA)

STEP 4: Clinical Utility Assessment
- ROC curves (sensitivity, specificity)
- Survival analysis (if outcome data available)
- Clinical decision curve analysis

RETURN:
- Top 10 biomarker candidates
- Validation plan with sample size calculations
- Clinical utility assessment framework

Use Case: Companion diagnostic development, patient stratification


5. Patient Stratification for Clinical Trial

Assess PatientOne's eligibility for precision oncology clinical trials:

STEP 1: Molecular Profile Summary
- Genomic: TP53 mutant, BRCA1 germline
- Pathways: PI3K/AKT/mTOR activated, DNA repair deficient
- Spatial: Immune-excluded tumor microenvironment
- Resistance: Platinum-resistant

STEP 2: Trial Matching
- Search ClinicalTrials.gov for:
  * BRCA-mutated ovarian cancer trials
  * PI3K/AKT/mTOR inhibitor trials
  * PARP inhibitor combinations
  * Immunotherapy + targeted therapy combinations
  * Bispecific antibody trials (PD-1xCTLA-4, MUC16xCD3, MSLNxCD47)
  * Cancer vaccine trials (IGFBP-2, neoantigen)
  * TIL (tumor-infiltrating lymphocyte) adoptive cell therapy trials

NOTE: STEP 2 uses the ClinicalTrials.gov connector (external).
See [Connect External MCP](../../for-researchers/CONNECT_EXTERNAL_MCP.md) for setup.

STEP 3: Eligibility Assessment
- Check inclusion criteria:
  * Stage IV HGSOC ✓
  * BRCA germline variant ✓
  * Platinum-resistant ✓
  * ECOG performance status (assume 0-1) ✓

STEP 4: Prioritization
- Rank trials by:
  * Molecular match score (how well profile fits)
  * Distance from patient location
  * Enrollment status (actively recruiting)

RETURN:
- Top 5 matching clinical trials
- Eligibility summary for each
- Contact information for trial coordinators

Use Case: Clinical trial enrollment, precision treatment access


5b. Immunotherapy Research Workflow for Cold Tumors (30 minutes)

Research next-generation immunotherapy candidates for PatientOne (PAT001-OVC-2025),
a platinum-resistant Stage IV HGSOC patient with cold/immune-excluded TME.

Patient data: gs://sample-inputs-patientone/patient-data/PAT001-OVC-2025/

STEP 1: Characterize Immune Landscape (8 minutes)
- Use mcp-spatialtools to map immune markers across tissue regions:
  * CD8A, CD3E, CD4 (T-cell markers)
  * CD68 (macrophage marker)
  * FOXP3 (Treg marker, if available)
  * Quantify immune exclusion: ratio of immune cells in stroma vs. tumor core
- Use mcp-quantum-celltype-fidelity to assess T-cell states:
  * Exhaustion markers: PDCD1 (PD-1), LAG3, HAVCR2 (TIM3), TIGIT
  * Activation markers: GZMB, PRF1, IFNG (if available)
- Use mcp-deepcell to quantify CD8+ cell density from IF imaging:
  * Cells/mm² in tumor core vs. stroma vs. periphery
  * Confirm immune-excluded phenotype

STEP 2: Check Immunotherapy Target Expression (7 minutes)
- Use mcp-multiomics to check RNA/protein levels for:
  * Checkpoint ligands: CD274 (PD-L1), CTLA4, LAG3, HAVCR2, TIGIT
    → Determines cadonilimab (PD-1xCTLA-4) candidacy
  * BiTE targets: MUC16 expression level, CD3D/CD3E on T cells
    → Determines ubamatamab (MUC16xCD3) candidacy
  * Innate immune targets: MSLN (mesothelin), CD47, CD68
    → Determines NI-1801 (MSLNxCD47) candidacy
  * Vaccine targets: IGFBP2 expression level
    → Determines IGFBP-2 vaccine candidacy
- Use mcp-spatialtools to map target spatial distribution:
  * Are targets expressed in tumor core, stroma, or both?
  * Note: Not all targets may be in the 31-gene Visium panel — check multi-omics for missing genes

STEP 3: Predict Treatment Response (7 minutes)
- Use mcp-perturbation (GEARS) to predict gene expression changes for:
  * Dual PD-1/CTLA-4 blockade (cadonilimab scenario)
  * MUC16-directed T-cell activation (ubamatamab scenario)
  * CD47 blockade + macrophage activation (NI-1801 scenario)
- Use mcp-quantum-celltype-fidelity to predict:
  * T-cell reactivation potential (for checkpoint/BiTE approaches)
  * Cold→hot TME conversion likelihood

STEP 4: Search Clinical Trials and Literature (5 minutes)
NOTE: This step uses EXTERNAL MCP tools (not built into platform servers).
These are available in Claude Desktop and Claude Code as separate MCP servers.
- Use ClinicalTrials.gov MCP (external) to search for:
  * Cadonilimab + ovarian cancer trials (e.g., NCT06560112)
  * NI-1801 MSLN-targeting trials (e.g., NCT05403554)
  * Ubamatamab MUC16xCD3 trials (e.g., NCT03564340, NCT06787612)
  * IGFBP-2 vaccine trials (e.g., NCT01322802)
  * TIL therapy in ovarian cancer (e.g., NCT01174121)
- Use PubMed MCP (external) to search for:
  * Latest efficacy data for each candidate
  * Cold tumor immunotherapy conversion strategies
  * HGSOC immunotherapy biomarker studies

STEP 5: Rank Candidates for PatientOne (3 minutes)
- Score each candidate on:
  * Cold TME suitability (can it work without T-cell infiltration?)
  * Target expression in PatientOne (confirmed from Steps 1-2)
  * Predicted response (from Step 3)
  * Clinical evidence strength (from Step 4)
  * Trial accessibility (recruiting, location, eligibility match)
- Generate ranked recommendation with rationale

RETURN: Structured immunotherapy research report with:
- Immune landscape characterization (hot/cold/excluded classification)
- Target expression summary table (present/absent/unknown for each target)
- Predicted response for each candidate (GEARS results)
- Top 5 clinical trials with NCT IDs and eligibility match
- Ranked immunotherapy recommendations (1-5) with evidence levels
- Limitations noted (gene panel coverage, synthetic data caveats)

Expected Results:

  • Immune landscape: Cold/immune-excluded (CD8+ low in tumor, high in stroma)
  • Top candidate: Ubamatamab (MUC16/CA-125 highly expressed, largest trial program)
  • Runner-up: NI-1801 (bypasses T-cell requirement via innate immunity)
  • Trials: 5+ recruiting trials with NCT IDs matching PatientOne's profile

Time: 30 minutes Cost: ~$2-5 (compute + tokens) Use Case: Immunotherapy research for cold/immune-excluded tumors, clinical trial matching for platinum-resistant HGSOC

Reference: See Immunotherapy Reference for detailed candidate profiles and decision framework.


Educational Workflows

6. PatientOne Teaching Case (25 minutes, $0.32)

DRY_RUN MODE: Complete PatientOne analysis for classroom demonstration.

STEP 1: Clinical Review (3 min)
- Review FHIR data (demographics, diagnosis, treatment)
- Identify key clinical questions

STEP 2: Genomic Analysis (5 min)
- Analyze VCF for TP53, BRCA1 variants
- Interpret clinical significance

STEP 3: Multi-Omics Quick Analysis (7 min)
- Load RNA, protein data
- Run differential expression
- Pathway enrichment (top 3 pathways)

STEP 4: Spatial Overview (7 min)
- Load Visium data
- Spatial DE and pathway enrichment
- Visualize on tissue

STEP 5: Treatment Summary (3 min)
- Synthesize findings
- Generate 3 treatment recommendations

RETURN: Student-friendly report (suitable for presentation)

Use Case: Undergraduate/graduate courses, workshops


Troubleshooting Workflows

7. Diagnostic Workflow for Failed Analysis

If previous analysis failed, run diagnostic workflow:

STEP 1: Data Availability Check
- List all available files for patient
- Verify file formats (VCF valid? CSV readable?)
- Check file sizes (empty files?)

STEP 2: Server Status Check
- Test each MCP server individually
- Check which tools are available
- Verify DRY_RUN mode status

STEP 3: Parameter Validation
- Verify patient ID is correct
- Check file paths exist
- Confirm thresholds are reasonable (FDR < 0.05 standard)

STEP 4: Incremental Analysis
- Start with simplest analysis (load data)
- Add complexity step by step
- Identify where failure occurs

STEP 5: Error Interpretation
- Parse error messages for actionable info
- Check common issues:
  * File not found → path typo
  * No significant results → threshold too strict
  * Server timeout → analysis too complex, break into steps

RETURN: Diagnostic report with recommended fixes

Use Case: Debugging, user support


Validation Workflows

8. Cross-Platform Validation

Validate PatientOne results against external data:

STEP 1: Internal Validation
- Compare RNA-seq to protein (concordance check)
- Spatial vs. bulk RNA (regional vs. average)
- Multiple pathway databases (KEGG, Hallmark, GO)

STEP 2: Literature Validation
- Top findings (TP53, BRCA1, PI3K pathway)
- Compare to published HGSOC studies
- Check expected prevalence (TP53 in 96% of HGSOC)

TIP: Use PubMed and bioRxiv connectors to search for validation studies
programmatically rather than manually. See [Connect External MCP](../../for-researchers/CONNECT_EXTERNAL_MCP.md).

STEP 3: Cohort Validation
- Query TCGA ovarian cancer cohort (mcp-mocktcga)
  TIP: Use cBioPortal connector for live TCGA-OV mutation and expression data.
  See [Connect External MCP](../../for-researchers/CONNECT_EXTERNAL_MCP.md).
- Compare PatientOne pathway activation to cohort
- Identify if patient is typical or outlier

STEP 4: Cross-Method Validation
- Confirm protein with IHC (if available)
- Validate spatial patterns with H&E histology
- Cross-check variants with orthogonal sequencing

RETURN: Validation report (concordance scores, literature support, cohort comparison)

Use Case: Quality assurance, publication preparation


Time-Saving Workflows

9. Quick Screening Workflow (10 minutes)

Quick PatientOne screening for triaging:

STEP 1: Genomic Quick Screen (3 min)
- Load VCF, check key genes only: TP53, BRCA1, BRCA2, PIK3CA
- Flag: Pathogenic variants YES/NO

STEP 2: Pathway Quick Screen (4 min)
- Load RNA data only (skip protein/phospho)
- Run pathway enrichment (KEGG only, top 5)
- Flag: Druggable pathways YES/NO

STEP 3: Spatial Quick Look (3 min)
- Load spatial data
- Cell type deconvolution (tumor % , immune %)
- Flag: Immune-infiltrated YES/NO

RETURN: Triage decision (High priority / Standard / Low priority for detailed analysis)

Use Case: High-throughput screening, resource allocation


10. Update Analysis with New Data

PatientOne has new post-treatment biopsy. Update analysis:

STEP 1: Load New Data
- New VCF (check for new mutations)
- New spatial data (compare tumor evolution)

STEP 2: Comparative Analysis
- Pre-treatment vs. Post-treatment
- What changed? (new mutations, pathway shifts, spatial reorganization)

STEP 3: Resistance Analysis
- Identify acquired resistance mechanisms
- Check for target loss, pathway bypass, microenvironment changes

STEP 4: Revised Treatment Plan
- Update recommendations based on new molecular state
- Suggest next-line therapies

RETURN: Updated report (changes highlighted, revised recommendations)

Use Case: Longitudinal monitoring, treatment adaptation


Quality Metrics

For any workflow, track:

  • Time to completion (target: < 60 min for comprehensive)
  • Cost (target: < $100)
  • Number of significant findings
  • Clinical actionability (% findings with drug implications)
  • Reproducibility (re-run should give identical results)

Related Prompts:


Last Updated: 2026-01-14