Submit
All skills
Skillintermediate

Real-World Scientific Examples

This document provides comprehensive, practical examples demonstrating how to combine Claude Scientific Skills to solve real scientific problems across multiple domains.

Claude Code Knowledge Pack7/10/2026

Overview

Real-World Scientific Examples

This document provides comprehensive, practical examples demonstrating how to combine Claude Scientific Skills to solve real scientific problems across multiple domains.

📋 Table of Contents

  1. Drug Discovery & Medicinal Chemistry
  2. Cancer Genomics & Precision Medicine
  3. Single-Cell Transcriptomics
  4. Protein Structure & Function
  5. Chemical Safety & Toxicology
  6. Clinical Trial Analysis
  7. Metabolomics & Systems Biology
  8. Materials Science & Chemistry
  9. Digital Pathology
  10. Lab Automation & Protocol Design
  11. Agricultural Genomics
  12. Neuroscience & Brain Imaging
  13. Environmental Microbiology
  14. Infectious Disease Research
  15. Multi-Omics Integration
  16. Computational Chemistry & Synthesis
  17. Clinical Research & Real-World Evidence
  18. Experimental Physics & Data Analysis
  19. Chemical Engineering & Process Optimization
  20. Scientific Illustration & Visual Communication
  21. Quantum Computing for Chemistry
  22. Research Grant Writing
  23. Flow Cytometry & Immunophenotyping

Drug Discovery & Medicinal Chemistry

Example 1: Discovery of Novel EGFR Inhibitors for Lung Cancer

Objective: Identify novel small molecule inhibitors of EGFR with improved properties compared to existing drugs.

Skills Used:

  • database-lookup - Query ChEMBL, PubChem, COSMIC, AlphaFold DB
  • paper-lookup - Search PubMed for literature
  • rdkit - Analyze molecular properties
  • datamol - Generate analogs
  • medchem - Medicinal chemistry filters
  • molfeat - Molecular featurization
  • diffdock - Molecular docking
  • deepchem - Property prediction
  • torchdrug - Graph neural networks for molecules
  • scientific-visualization - Create figures
  • clinical-reports - Generate PDF reports

Workflow:

# Always use available 'skills' when possible. Keep the output organized.

Step 1: Query ChEMBL for known EGFR inhibitors with high potency
- Search for compounds targeting EGFR (CHEMBL203)
- Filter: IC50 < 50 nM, pChEMBL value > 7
- Extract SMILES strings and activity data
- Export to DataFrame for analysis

Step 2: Analyze structure-activity relationships
- Load compounds into RDKit
- Calculate molecular descriptors (MW, LogP, TPSA, HBD, HBA)
- Generate Morgan fingerprints (radius=2, 2048 bits)
- Perform hierarchical clustering to identify scaffolds
- Visualize top scaffolds with activity annotations

Step 3: Identify resistance mutations from COSMIC
- Query COSMIC for EGFR mutations in lung cancer
- Focus on gatekeeper mutations (T790M, C797S)
- Extract mutation frequencies and clinical significance
- Cross-reference with literature in PubMed

Step 4: Retrieve EGFR structure from AlphaFold
- Download AlphaFold prediction for EGFR kinase domain
- Alternatively, use experimental structure from PDB (if available)
- Prepare structure for docking (add hydrogens, optimize)

Step 5: Generate novel analogs using datamol
- Select top 5 scaffolds from ChEMBL analysis
- Use scaffold decoration to generate 100 analogs per scaffold
- Apply Lipinski's Rule of Five filtering
- Ensure synthetic accessibility (SA score < 4)
- Check for PAINS and unwanted substructures

Step 6: Predict properties with DeepChem
- Train graph convolutional model on ChEMBL EGFR data
- Predict pIC50 for generated analogs
- Predict ADMET properties (solubility, permeability, hERG)
- Rank candidates by predicted potency and drug-likeness

Step 7: Virtual screening with DiffDock
- Perform molecular docking on top 50 candidates
- Dock into wild-type EGFR and T790M mutant
- Calculate binding energies and interaction patterns
- Identify compounds with favorable binding to both forms

Step 8: Search PubChem for commercial availability
- Query PubChem for top 10 candidates by InChI key
- Check supplier information and purchasing options
- Identify close analogs if exact matches unavailable

Step 9: Literature validation with PubMed
- Search for any prior art on top scaffolds
- Query: "[scaffold_name] AND EGFR AND inhibitor"
- Summarize relevant findings and potential liabilities

Step 10: Create comprehensive report
- Generate 2D structure visualizations of top hits
- Create scatter plots: MW vs LogP, TPSA vs potency
- Produce binding pose figures for top 3 compounds
- Generate table comparing properties to approved drugs (gefitinib, erlotinib)
- Write scientific summary with methodology, results, and recommendations
- Export to PDF with proper citations

Expected Output: 
- Ranked list of 10-20 novel EGFR inhibitor candidates
- Predicted activity and ADMET properties
- Docking poses and binding analysis
- Comprehensive scientific report with publication-quality figures

Example 2: Drug Repurposing for Rare Diseases

Objective: Identify FDA-approved drugs that could be repurposed for treating a rare metabolic disorder.

Skills Used:

  • database-lookup - Query DrugBank, Open Targets, STRING, KEGG, Reactome, ClinicalTrials.gov, FDA
  • paper-lookup - Search OpenAlex, bioRxiv, PubMed
  • networkx - Network analysis
  • bioservices - Biological database queries
  • literature-review - Systematic review

Workflow:

Step 1: Define disease pathway
- Query KEGG and Reactome for disease-associated pathways
- Identify key proteins and enzymes involved
- Map upstream and downstream pathway components

Step 2: Find protein-protein interactions
- Query STRING database for interaction partners
- Build protein interaction network around key disease proteins
- Identify hub proteins and bottlenecks using NetworkX
- Calculate centrality metrics (betweenness, closeness)

Step 3: Query Open Targets for druggable targets
- Search for targets associated with disease phenotype
- Filter by clinical precedence and tractability
- Prioritize targets with existing approved drugs

Step 4: Search DrugBank for drugs targeting identified proteins
- Query for approved drugs and their targets
- Filter by mechanism of action relevant to disease
- Retrieve drug properties and safety information

Step 5: Query FDA databases for safety profiles
- Check FDA adverse event database (FAERS)
- Review drug labels and black box warnings
- Assess risk-benefit for rare disease population

Step 6: Search ClinicalTrials.gov for prior repurposing attempts
- Query for disease name + drug names
- Check for failed trials (and reasons for failure)
- Identify ongoing trials that may compete

Step 7: Perform pathway enrichment analysis
- Map drug targets to disease pathways
- Calculate enrichment scores with Reactome
- Identify drugs affecting multiple pathway nodes

Step 8: Conduct systematic literature review
- Search PubMed for drug name + disease associations
- Include bioRxiv for recent unpublished findings
- Document any case reports or off-label use
- Use literature-review skill to generate comprehensive review

Step 9: Prioritize candidates
- Rank by: pathway relevance, safety profile, existing evidence
- Consider factors: oral availability, blood-brain barrier penetration
- Assess commercial viability and patent status

Step 10: Generate repurposing report
- Create network visualization of drug-target-pathway relationships
- Generate comparison table of top 5 candidates
- Write detailed rationale for each candidate
- Include mechanism of action diagrams
- Provide recommendations for preclinical validation
- Format as professional PDF with citations

Expected Output:
- Ranked list of 5-10 repurposing candidates
- Network analysis of drug-target-disease relationships
- Safety and efficacy evidence summary
- Repurposing strategy report with next steps

Cancer Genomics & Precision Medicine

Example 3: Clinical Variant Interpretation Pipeline

Objective: Analyze a patient's tumor sequencing data to identify actionable mutations and therapeutic recommendations.

Skills Used:

  • database-lookup - Query Ensembl, ClinVar, COSMIC, NCBI Gene, UniProt, ClinPGx, DrugBank, ClinicalTrials.gov, Open Targets
  • paper-lookup - Search PubMed for literature evidence
  • pysam - Parse VCF files
  • gget - Unified gene/protein data retrieval
  • clinical-reports - Generate clinical report PDF

Workflow:

Step 1: Parse and filter VCF file
- Use pysam to read tumor VCF
- Filter for high-quality variants (QUAL > 30, DP > 20)
- Extract variant positions, alleles, and VAF (variant allele frequency)
- Separate SNVs, indels, and structural variants

Step 2: Annotate variants with Ensembl
- Query Ensembl VEP API for functional consequences
- Classify variants: missense, nonsense, frameshift, splice site
- Extract transcript information and protein changes
- Identify canonical transcripts for each gene

Step 3: Query ClinVar for known pathogenic variants
- Search ClinVar by genomic coordinates
- Extract clinical significance classifications
- Note conflicting interpretations and review status
- Prioritize variants with "Pathogenic" or "Likely Pathogenic" labels

Step 4: Query COSMIC for somatic cancer mutations
- Search COSMIC for each variant
- Extract mutation frequency across cancer types
- Identify hotspot mutations (high recurrence)
- Note drug resistance mutations

Step 5: Retrieve gene information from NCBI Gene
- Get detailed gene descriptions
- Extract associated phenotypes and diseases
- Identify oncogene vs tumor suppressor classification
- Note gene function and biological pathways

Step 6: Assess protein-level impact with UniProt
- Query UniProt for protein domain information
- Map variants to functional domains (kinase domain, binding site)
- Check if variant affects active sites or protein stability
- Retrieve post-translational modification sites

Step 7: Search DrugBank for targetable alterations
- Query for drugs targeting mutated genes
- Filter for FDA-approved and investigational drugs
- Extract mechanism of action and indications
- Prioritize variants with approved targeted therapies

Step 8: Query Open Targets for target-disease associations
- Validate therapeutic hypotheses
- Assess target tractability scores
- Review clinical precedence for each gene-disease pair

Step 9: Search ClinicalTrials.gov for matching trials
- Build query with: cancer type + gene names + variants
- Filter for: recruiting status, phase II/III trials
- Extract trial eligibility criteria
- Note geographic locations and contact information

Step 10: Literature search for clinical evidence
- PubMed query: "[gene] AND [variant] AND [cancer type]"
- Focus on: case reports, clinical outcomes, resistance mechanisms
- Extract relevant prognostic or predictive information

Step 11: Classify variants by actionability
Tier 1: FDA-approved therapy for this variant
Tier 2: Clinical trial available for this variant
Tier 3: Therapy approved for variant in different cancer
Tier 4: Biological evidence but no approved therapy

Step 12: Generate clinical genomics report
- Executive summary of key findings
- Table of actionable variants with evidence levels
- Therapeutic recommendations with supporting evidence
- Clinical trial options with eligibility information
- Prognostic implications based on mutation profile
- References to guidelines (NCCN, ESMO, AMP/ASCO/CAP)
- Generate professional PDF using clinical-reports skill

Expected Output:
- Annotated variant list with clinical significance
- Tiered list of actionable mutations
- Therapeutic recommendations with evidence levels
- Matching clinical trials
- Comprehensive clinical genomics report (PDF)

Example 4: Cancer Subtype Classification from Gene Expression

Objective: Classify breast cancer subtypes using RNA-seq data and identify subtype-specific therapeutic vulnerabilities.

Skills Used:

  • database-lookup - Query NCBI Gene, Reactome, Open Targets
  • paper-lookup - Search PubMed for literature validation
  • pydeseq2 - Differential expression
  • scanpy - Clustering and visualization
  • scikit-learn - Machine learning classification
  • gget - Gene data retrieval
  • matplotlib - Visualization
  • seaborn - Heatmaps
  • plotly - Interactive visualization
  • scikit-survival - Survival analysis

Workflow:

Step 1: Load and preprocess RNA-seq data
- Load count matrix (genes × samples)
- Filter low-expression genes (mean counts < 10)
- Normalize with DESeq2 size factors
- Apply variance-stabilizing transformation (VST)

Step 2: Classify samples using PAM50 genes
- Query NCBI Gene for PAM50 classifier gene list
- Extract expression values for PAM50 genes
- Train Random Forest classifier on labeled training data
- Predict subtypes: Luminal A, Luminal B, HER2+, Basal, Normal-like
- Validate with published markers (ESR1, PGR, ERBB2, MKI67)

Step 3: Perform differential expression for each subtype
- Use PyDESeq2 to compare each subtype vs all others
- Apply multiple testing correction (FDR < 0.05)
- Filter by log2 fold change (|LFC| > 1.5)
- Identify subtype-specific signature genes

Step 4: Annotate differentially expressed genes
- Query NCBI Gene for detailed annotations
- Classify as oncogene, tumor suppressor, or other
- Extract biological process and molecular function terms

Step 5: Pathway enrichment analysis
- Submit gene lists to Reactome API
- Identify enriched pathways for each subtype (p < 0.01)
- Focus on druggable pathways (kinase signaling, metabolism)
- Compare pathway profiles across subtypes

Step 6: Identify therapeutic targets with Open Targets
- Query Open Targets for each upregulated gene
- Filter by tractability score > 5
- Prioritize targets with clinical precedence
- Extract associated drugs and development phase

Step 7: Create comprehensive visualization
- Generate UMAP projection of all samples colored by subtype
- Create heatmap of PAM50 genes across subtypes
- Produce volcano plots for each subtype comparison
- Generate pathway enrichment dot plots
- Create drug target-pathway network diagrams

Step 8: Literature validation
- Search PubMed for each predicted therapeutic target
- Query: "[gene] AND [subtype] AND breast cancer AND therapy"
- Summarize clinical evidence and ongoing trials
- Note any resistance mechanisms reported

Step 9: Generate subtype-specific recommendations
For each subtype:
- List top 5 differentially expressed