mims-harvard/tooluniverse-infectious-disease

COMPUTE, DON'T DESCRIBE

When analysis requires computation (statistics, data processing, scoring, enrichment), write and run Python code via Bash. Don't describe what you would do — execute it and report actual results. Use ToolUniverse tools to retrieve data, then Python (pandas, scipy, statsmodels, matplotlib) to analyze it.

Infectious Disease Outbreak Intelligence

Rapid response system for emerging pathogens using taxonomy analysis, target identification, structure prediction, and computational drug repurposing.

KEY PRINCIPLES:

1. Speed is critical - Optimize for rapid actionable intelligence
2. Target essential proteins - Focus on conserved, essential viral/bacterial proteins
3. Leverage existing drugs - Prioritize FDA-approved compounds for repurposing
4. Structure-guided - Use NvidiaNIM for rapid structure prediction and docking
5. Evidence-graded - Grade repurposing candidates by evidence strength
6. Actionable output - Prioritized drug candidates with rationale
7. English-first queries - Always use English terms in tool calls; respond in user's language

REASONING STRATEGY — Start Here: Start with pathogen identification: What type of organism? (virus, bacteria, fungus, parasite). Then ask:

• What are the essential proteins? (required for replication or viability — cannot be mutated away)
• Which are surface-exposed? (accessible to drugs and antibodies)
• Which are conserved across strains? (targeting conserved regions prevents resistance escape) These three questions define your drug targets and vaccine candidates. Organisms in the same genus share targets — look up drug precedent for related pathogens before predicting from scratch.

LOOK UP DON'T GUESS: Never assume a pathogen's taxonomy, genome size, or protein function. Always call BVBRC_search_taxonomy or UniProt_search first. Even well-known pathogens have strains with different drug susceptibility profiles — look up the specific strain when known.

---

When to Use

Apply when user asks:

• "New pathogen detected - what drugs might work?"
• "Emerging virus [X] - therapeutic options?"
• "Drug repurposing candidates for [pathogen]"
• "What do we know about [novel coronavirus/bacteria]?"
• "Essential targets in [pathogen] for drug development"
• "Can we repurpose [drug] against [pathogen]?"

---

Critical Workflow Requirements

1. Report-First Approach (MANDATORY)

1. Create [PATHOGEN]_outbreak_intelligence.md FIRST with section headers
2. Progressively update as data is gathered
3. Output separate files: [PATHOGEN]_drug_candidates.csv, [PATHOGEN]_target_proteins.csv

2. Citation Requirements (MANDATORY)

Every finding must have inline source attribution:

### Target: RNA-dependent RNA polymerase (RdRp)
- **UniProt**: P0DTD1 (NSP12)
- **Essentiality**: Required for replication
*Source: UniProt via `UniProt_search`, literature review*

---

Phase 0: Tool Verification

Known Parameter Corrections

| Tool | WRONG Parameter | CORRECT Parameter | |------|-----------------|-------------------| | NCBIDatasets_get_taxonomy | name | tax_id (integer) or use BVBRC_search_taxonomy for keyword search | | UniProt_search | name | query | | ChEMBL_search_targets | query, target | pref_name__contains (substring match) | | get_diffdock_info | protein_file | protein (content) | | drugbank_full_search | (may fail) | Use drugbank_vocab_search as primary DrugBank lookup |

PubMed tip: Use sort="relevance" (default) not sort="pub_date" — date-sorted queries can return empty for narrow topics. Tool name: PubMed_search_articles. FDA labels: Use FDA_get_drug_label_info_by_field_value with targeted return_fields to avoid oversized responses from OpenFDA_search_drug_labels.

---

Workflow Overview

Phase 1: Pathogen Identification
├── Taxonomic classification (NCBI Taxonomy)
├── Closest relatives (for knowledge transfer)
├── Genome/proteome availability
└── OUTPUT: Pathogen profile
    |
Phase 2: Target Identification
├── Essential genes/proteins (UniProt)
├── Conservation across strains
├── Druggability assessment (ChEMBL)
└── OUTPUT: Prioritized target list (scored by essentiality/conservation/druggability/precedent)
    |
Phase 3: Structure Prediction (NvidiaNIM)
├── AlphaFold2/ESMFold for targets
├── Binding site identification
├── Quality assessment (pLDDT)
└── OUTPUT: Target structures (docking-ready if pLDDT > 70)
    |
Phase 4: Drug Repurposing Screen
├── Approved drugs for related pathogens (ChEMBL)
├── Broad-spectrum antivirals/antibiotics
├── Docking screen (get_diffdock_info)
└── OUTPUT: Ranked candidate drugs
    |
Phase 4.5: Pathway Analysis
├── KEGG: Pathogen metabolism pathways
├── Essential metabolic targets
├── Host-pathogen interaction pathways
└── OUTPUT: Pathway-based drug targets
    |
Phase 5: Literature Intelligence
├── PubMed: Published outbreak reports
├── BioRxiv/MedRxiv: Recent preprints (CRITICAL for outbreaks)
├── ArXiv: Computational/ML preprints
├── OpenAlex: Citation tracking
├── ClinicalTrials.gov: Active trials
└── OUTPUT: Evidence synthesis
    |
Phase 6: Report Synthesis
├── Top drug candidates with evidence grades
├── Clinical trial opportunities
├── Recommended immediate actions
└── OUTPUT: Final report

---

Phase Summaries

Phase 1: Pathogen Identification

Classify via NCBI Taxonomy (query param). Identify related pathogens with existing drugs for knowledge transfer. Determine genome/proteome availability.

Knowledge transfer principle: Drugs effective against related pathogens are the highest-priority repurposing candidates. A protease inhibitor for SARS-CoV-1 is immediately relevant to SARS-CoV-2. Look up the related pathogen's approved drugs in ChEMBL before generating candidates from first principles.

Phase 2: Target Identification

Search UniProt for pathogen proteins (reviewed). Check ChEMBL for drug precedent. Score targets by: Essentiality (30%), Conservation (25%), Druggability (25%), Drug precedent (20%). Aim for 5+ targets.

Phase 3: Structure Prediction

Use NvidiaNIM AlphaFold2 for top 3 targets. Assess pLDDT confidence. Only dock structures with pLDDT > 70 (active site > 90 preferred). Fallback: alphafold_get_prediction or ESMFold_predict_structure.

Phase 4: Drug Repurposing Screen

Source candidates from: related pathogen drugs, broad-spectrum antivirals, target class drugs (DGIdb). Dock top 20+ candidates via get_diffdock_info. Rank by docking score and evidence tier.

Phase 4.5: Pathway Analysis

Use KEGG to identify essential metabolic pathways. Map host-pathogen interaction points. Identify pathway-based drug targets beyond direct protein inhibition.

Phase 5: Literature Intelligence

Search PubMed (peer-reviewed), BioRxiv/MedRxiv (preprints - critical for outbreaks), ArXiv (computational), ClinicalTrials.gov (active trials). Track citations via OpenAlex. Note: preprints are NOT peer-reviewed.

Phase 6: Report Synthesis

Aggregate all findings into final report. Grade every candidate. Provide 3+ immediate actions, clinical trial opportunities, and research priorities.

---

Evidence Grading

| Tier | Symbol | Criteria | Example | |------|--------|----------|---------| | T1 | [T1] | FDA approved for this pathogen | Remdesivir for COVID | | T2 | [T2] | Clinical trial evidence OR approved for related pathogen | Favipiravir | | T3 | [T3] | In vitro activity OR strong docking + mechanism | Sofosbuvir | | T4 | [T4] | Computational prediction only | Novel docking hits |

---

Completeness Checklist

Phase 1: Pathogen ID

• [ ] Taxonomic classification complete
• [ ] Related pathogens identified
• [ ] Genome/proteome availability noted

Phase 2: Targets

• [ ] 5+ targets identified
• [ ] Essentiality documented
• [ ] Conservation assessed
• [ ] Drug precedent checked

Phase 3: Structures

• [ ] Structures predicted for top 3 targets
• [ ] pLDDT confidence reported
• [ ] Binding sites identified

Phase 4: Drug Screen

• [ ] 20+ candidates screened
• [ ] FDA-approved drugs prioritized
• [ ] Docking scores reported
• [ ] Top 5 candidates detailed

Phase 5: Literature

• [ ] Recent papers summarized
• [ ] Active trials listed
• [ ] Resistance data noted

Phase 6: Recommendations

• [ ] 3+ immediate actions
• [ ] Clinical trial opportunities
• [ ] Research priorities

---

Fallback Chains

| Primary Tool | Fallback 1 | Fallback 2 | |--------------|------------|------------| | NvidiaNIM_alphafold2 (requires NVIDIA_API_KEY env var; free key at build.nvidia.com) | alphafold_get_prediction (AlphaFold DB by UniProt) | ESMFold_predict_structure | | get_diffdock_info | NvidiaNIM_boltz2 (requires NVIDIA_API_KEY env var; free key at build.nvidia.com) | Manual docking | | NCBIDatasets_suggest_taxonomy | UniProtTaxonomy_get_taxon | Manual classification | | ChEMBL_search_drugs | drugbank_vocab_search | PubChem bioassays |

---

References

| File | Contents | |------|----------| | TOOLS_REFERENCE.md | Complete tool documentation | | phase_details.md | Detailed code examples and procedures for each phase | | report_template.md | Report template with section headers, checklist, and evidence grading | | CHECKLIST.md | Pre-delivery verification checklist (quality, citations, docking) | | EXAMPLES.md | Full worked examples (coronavirus, CRKP, limited-info scenarios) |