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

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.

Genome assembly availability and QC: After classifying the pathogen, use NCBIDatasets_list_genomes_by_taxon (params taxon as tax_id, limit, reference_only) to find the reference genome, NCBIDatasets_get_genome_assembly (param accession, e.g. "GCF_000005845.2") for assembly metrics (length, N50, GC%, contig/chromosome counts), and NCBIDatasets_get_sequence_reports (param accession) to map replicons (chromosomes/plasmids with RefSeq/GenBank accessions). For the full assembly-QC-to-characterization workflow, see the tooluniverse-microbial-genome-characterization skill.

Open pathogen genomic surveillance: For the priority pathogens covered by Pathoplexus (west-nile, ebola-zaire, ebola-sudan, cchf, mpox), use Pathoplexus_count_sequences (params organism, group_by e.g. geoLocCountry or lineage) to gauge sequencing volume and geographic/lineage spread, and Pathoplexus_get_mutations (params organism, min_proportion e.g. 0.95) to pull characteristic high-prevalence mutations for the circulating population. Use early to quantify outbreak footprint and flag conserved mutations before target selection.

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

---

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

---

References

Infectious Disease Outbreak Intelligence Checklist

Pre-delivery verification checklist for outbreak intelligence reports.

Report Quality Checklist

Structure & Format

• [ ] Report file created: [PATHOGEN]_outbreak_intelligence.md
• [ ] All 8 main sections present
• [ ] Executive summary completed (not [Analyzing...])
• [ ] Data sources section populated

Phase 1: Pathogen Identification

• [ ] Scientific name documented
• [ ] Taxonomy ID (NCBI) provided
• [ ] Pathogen type classified (virus/bacteria/fungus/parasite)
• [ ] Family and genus identified
• [ ] Full taxonomic lineage shown
• [ ] Related pathogens with drug precedent identified
• [ ] Genome/proteome reference noted

Phase 2: Target Identification

• [ ] ≥5 druggable targets identified
• [ ] Each target has UniProt accession
• [ ] Protein function described
• [ ] Essentiality justified
• [ ] Conservation across strains noted
• [ ] Drug precedent checked (related pathogens)
• [ ] Targets ranked by druggability score
• [ ] Top 3 targets have detailed profiles

Phase 3: Structure Prediction

• [ ] NVIDIA_API_KEY availability documented
• [ ] Structures predicted for top 3 targets (minimum)
• [ ] Method stated (AlphaFold2/ESMFold)
• [ ] Mean pLDDT reported for each structure
• [ ] pLDDT distribution by region included
• [ ] Binding site regions identified
• [ ] Active site residue confidence documented
• [ ] Docking suitability assessed

Phase 4: Drug Repurposing Screen

• [ ] ≥20 candidate drugs identified
• [ ] Sources documented (related pathogen, broad-spectrum, target class)
• [ ] FDA approval status for each drug
• [ ] Docking performed for top candidates
• [ ] Docking scores reported with confidence
• [ ] Top 5-10 candidates ranked
• [ ] Mechanism of action described
• [ ] Clinical evidence level assigned

Phase 5: Literature Intelligence

• [ ] Recent publications searched (<6 months)
• [ ] Key findings summarized
• [ ] Active clinical trials listed
• [ ] Resistance data noted (if available)
• [ ] Previous treatment outcomes summarized

Phase 6: Recommendations

• [ ] ≥3 immediate actions listed
• [ ] Clinical trial opportunities identified
• [ ] Research priorities outlined
• [ ] Timeframe/urgency noted

---

Citation Requirements

Every Section Must Include

• [ ] Source database name
• [ ] Tool used (in backticks)
• [ ] Specific identifiers (UniProt, NCBI, NCT, etc.)

Format Examples

*Source: NCBI Taxonomy via `NCBIDatasets_suggest_taxonomy` (TaxID: 2697049)*
*Source: UniProt via `UniProt_search` (P0DTD1)*
*Source: NVIDIA NIM via `NvidiaNIM_alphafold2` (pLDDT: 91.2)*
*Source: NVIDIA NIM via `NvidiaNIM_diffdock` (score: 0.92)*
*Source: ChEMBL via `ChEMBL_search_drugs`*

---

Evidence Grading

All Drug Candidates Must Have

• [ ] Evidence tier assigned (★★★ to ☆☆☆)
• [ ] Docking score (if docked)
• [ ] Clinical evidence level
• [ ] Rationale for ranking

Tier Definitions

---

Quantified Minimums

---

Pathogen-Specific Checks

Viral Pathogens

• [ ] Genome type noted (DNA/RNA, ss/ds, +/-)
• [ ] Polymerase identified as target
• [ ] Protease(s) identified
• [ ] Entry mechanism proteins noted
• [ ] Polymerase inhibitors screened
• [ ] Protease inhibitors screened

Bacterial Pathogens

• [ ] Gram stain classification
• [ ] Essential metabolic pathways identified
• [ ] Antibiotic resistance genes checked
• [ ] Cell wall synthesis targets noted
• [ ] Protein synthesis targets noted
• [ ] DNA replication targets noted

Drug-Resistant Pathogens

• [ ] Resistance mechanisms documented
• [ ] Drugs avoiding resistance prioritized
• [ ] Novel mechanism drugs highlighted
• [ ] Combination strategies considered

---

Docking Quality Checks

Structure Quality

• [ ] Mean pLDDT >70 for docking
• [ ] Active site pLDDT >80
• [ ] No major clashes in structure
• [ ] Binding pocket well-defined

Docking Results

• [ ] Reference drug docked for validation
• [ ] Score interpretation provided
• [ ] Pose quality assessed
• [ ] Binding mode plausible

---

Output Files

Required

• [ ] [PATHOGEN]_outbreak_intelligence.md - Main report

Recommended

• [ ] [PATHOGEN]_drug_candidates.csv - All candidates with scores
• [ ] [PATHOGEN]_target_proteins.csv - Target list with properties

CSV Column Requirements

drug_candidates.csv:

Drug_Name,ChEMBL_ID,Indication,FDA_Status,Docking_Score,Evidence_Tier,Mechanism

target_proteins.csv:

Target_Name,UniProt_ID,Function,Essentiality,Conservation,Druggability_Score,Drug_Precedent

---

Urgent Findings Protocol

If any of these found, flag prominently:

• [ ] No approved drugs available
• [ ] All existing drugs ineffective (docking)
• [ ] High mortality/transmission
• [ ] Drug resistance detected
• [ ] Novel pathogen (no related drugs)

Urgent Flag Format

⚠️ **URGENT: [Finding]** ⚠️

[Description of critical finding]

**Recommended Action**: [Immediate step]

---

Final Review

Before Delivery

• [ ] No [Analyzing...] placeholders remaining
• [ ] All tables properly formatted
• [ ] Executive summary synthesizes key findings
• [ ] Top drug candidate clearly stated
• [ ] Recommendations are actionable and specific
• [ ] Appropriate urgency conveyed

Common Issues to Avoid

• [ ] Targets without UniProt IDs
• [ ] Structures without pLDDT confidence
• [ ] Drugs without FDA status
• [ ] Docking without reference compound
• [ ] Recommendations without evidence support
• [ ] Missing pathogen classification

---

Speed Optimization

For outbreak scenarios, prioritize: 1. FDA-approved drugs first (fastest to deploy) 2. Drugs in phase 3 trials second 3. Novel candidates last (longest timeline)

Parallel Processing

• Run taxonomy AND protein searches in parallel
• Run multiple structure predictions in parallel
• Dock multiple candidates simultaneously

Infectious Disease Outbreak Intelligence - Examples

Example 1: Novel Coronavirus Outbreak

User Query

"New coronavirus detected in bats with potential human spillover risk. What drugs might work if it spreads?"

Expected Output

# Outbreak Intelligence Report: Novel Betacoronavirus

**Generated**: 2026-02-04 | **Status**: URGENT - Preparedness Assessment

---

## Executive Summary

This novel betacoronavirus shares key features with SARS-CoV-2, enabling rapid knowledge transfer. **Top drug candidates** based on structural similarity and docking:

1. **Nirmatrelvir (Paxlovid component)** ★★★ - Main protease inhibitor
2. **Remdesivir** ★★★ - RdRp inhibitor
3. **Molnupiravir** ★★☆ - Mutagenic nucleoside analog

**Recommended immediate actions**:
1. Stockpile nirmatrelvir/ritonavir (highest docking scores)
2. Prepare remdesivir supply chain
3. Sequence additional isolates for resistance surveillance

---

## 1. Pathogen Profile

### 1.1 Taxonomic Classification

| Property | Value |
|----------|-------|
| **Organism** | Novel betacoronavirus (provisional) |
| **Type** | RNA virus (+ssRNA) |
| **Family** | Coronaviridae |
| **Genus** | Betacoronavirus |
| **Subgenus** | Sarbecovirus (based on phylogeny) |
| **Closest relative** | SARS-CoV-2 (87% genome identity) |

### 1.2 Related Pathogens with Drug Precedent

| Pathogen | Identity | Approved Drugs | Transferability |
|----------|----------|----------------|-----------------|
| SARS-CoV-2 | 87% | Nirmatrelvir, Remdesivir, Molnupiravir | HIGH |
| SARS-CoV | 79% | (Remdesivir EUA) | HIGH |
| MERS-CoV | 52% | None approved | MEDIUM |
| HCoV-OC43 | 45% | None specific | LOW |

**Assessment**: High drug transferability expected from SARS-CoV-2 therapeutics.

*Source: NCBI Taxonomy, BLAST alignment*

---

## 2. Druggable Targets

### 2.1 Prioritized Target List

| Rank | Target | Gene | Function | Score | Drug Precedent |
|------|--------|------|----------|-------|----------------|
| 1 | Main protease (Mpro) | NSP5 | Polyprotein cleavage | 95 | Nirmatrelvir |
| 2 | RdRp (NSP12) | NSP12 | RNA replication | 92 | Remdesivir |
| 3 | Papain-like protease | NSP3 | Deubiquitination | 78 | GRL0617 |
| 4 | Spike protein | S | Host entry | 75 | Antibodies |
| 5 | Helicase | NSP13 | RNA unwinding | 68 | None approved |

### 2.2 Top Target: Main Protease (Mpro)

| Property | Value |
|----------|-------|
| **Gene** | NSP5 (polyprotein position 3264-3569) |
| **Length** | 306 amino acids |
| **Function** | Cleaves viral polyprotein at 11 sites |
| **Essentiality** | Absolute - no viral replication without Mpro |
| **Conservation** | 96% identity with SARS-CoV-2 Mpro |
| **Active site** | Cys145-His41 catalytic dyad (100% conserved) |

**Critical finding**: Active site 100% conserved → nirmatrelvir expected to work.

*Source: UniProt, sequence alignment*

---

## 3. Target Structures

### 3.1 Structure Prediction Results

| Target | Method | Length | Mean pLDDT | Docking Ready |
|--------|--------|--------|------------|---------------|
| Mpro | AlphaFold2 | 306 aa | 94.2 | ✓ Excellent |
| RdRp | AlphaFold2 | 932 aa | 91.8 | ✓ Yes |
| PLpro | AlphaFold2 | 315 aa | 87.3 | ✓ Yes |

### 3.2 Mpro Structure Quality

| Region | Residues | pLDDT | Functional Role |
|--------|----------|-------|-----------------|
| Domain I | 8-101 | 95.1 | Substrate binding |
| Domain II | 102-184 | 94.8 | Catalytic |
| Domain III | 201-303 | 92.4 | Dimerization |
| **Active site** | C145, H41 | **97.2** | Catalysis |
| S1 pocket | F140, H163 | 96.8 | Substrate specificity |

**Docking Assessment**: Structure highly suitable; active site pLDDT 97.2.

*Source: NVIDIA NIM via `NvidiaNIM_alphafold2`*

---

## 4. Drug Repurposing Screen

### 4.1 Candidate Sources

| Source | Candidates | FDA Approved |
|--------|------------|--------------|
| SARS-CoV-2 drugs | 5 | 3 |
| Broad-spectrum antivirals | 12 | 8 |
| Protease inhibitors | 8 | 6 |
| RdRp inhibitors | 6 | 4 |
| **Total unique** | **24** | **15** |

### 4.2 Docking Results - Mpro Target

| Rank | Drug | Original Indication | Docking Score | Evidence |
|------|------|---------------------|---------------|----------|
| 1 | **Nirmatrelvir** | COVID-19 | 0.94 | ★★★ FDA approved |
| 2 | **Ensitrelvir** | COVID-19 (Japan) | 0.91 | ★★☆ Approved Japan |
| 3 | Boceprevir | HCV | 0.82 | ★☆☆ In vitro only |
| 4 | Lopinavir | HIV | 0.78 | ★☆☆ Mixed results |
| 5 | GRL0617 | Research | 0.75 | ☆☆☆ Preclinical |

### 4.3 Docking Results - RdRp Target

| Rank | Drug | Docking Score | Evidence |
|------|------|---------------|----------|
| 1 | **Remdesivir** | 0.93 | ★★★ FDA approved |
| 2 | **Molnupiravir** | 0.87 | ★★★ FDA approved |
| 3 | Favipiravir | 0.84 | ★★☆ Approved some countries |
| 4 | Sofosbuvir | 0.81 | ★☆☆ In vitro active |
| 5 | Ribavirin | 0.72 | ★☆☆ Weak activity |

### 4.4 Top Candidate: Nirmatrelvir

| Property | Value |
|----------|-------|
| **Docking score** | 0.94 (excellent, >reference) |
| **Mechanism** | Covalent inhibitor of Mpro active site |
| **FDA status** | Approved (with ritonavir as Paxlovid) |
| **Key binding** | C145 (covalent), H163 (H-bond) |
| **Expected efficacy** | HIGH (conserved target) |

*Source: NVIDIA NIM via `NvidiaNIM_diffdock`, ChEMBL*

---

## 5. Literature Intelligence

### 5.1 Relevant Knowledge Base

| Topic | Key Insight | Source |
|-------|-------------|--------|
| Mpro inhibitors | Conserved active site across sarbecoviruses | PMID:35892324 |
| Resistance | E166V mutation reduces nirmatrelvir binding | PMID:36112789 |
| RdRp inhibitors | Remdesivir active against all tested coronaviruses | PMID:34567890 |

### 5.2 Surveillance Considerations

**Monitor for resistance mutations**:
- Mpro: E166V, H172Y (nirmatrelvir resistance)
- RdRp: F480L, V557L (remdesivir resistance)

*Source: PubMed literature review*

---

## 6. Recommendations

### 6.1 Immediate Actions (Preparedness)

| Priority | Action | Rationale |
|----------|--------|-----------|
| 1 | **Stockpile Paxlovid** | Highest docking score, FDA approved |
| 2 | **Secure remdesivir supply** | Backup, especially for severe cases |
| 3 | **Sequence surveillance** | Monitor for resistance mutations |
| 4 | **Prepare clinical protocols** | Ready if human cases occur |

### 6.2 If Human Spillover Occurs

1. **First-line**: Paxlovid (nirmatrelvir/ritonavir) within 5 days of symptoms
2. **Hospitalized**: Add remdesivir
3. **Resistance detected**: Consider molnupiravir or combination therapy

### 6.3 Research Priorities

1. Confirm in vitro activity of top candidates against isolate
2. Animal model efficacy studies
3. Develop resistance monitoring panel
4. Evaluate antibody cross-reactivity from COVID-19 vaccines

---

## 7. Data Gaps

| Gap | Impact | Mitigation |
|-----|--------|------------|
| In vitro confirmation | Docking is predictive | Priority testing needed |
| Resistance baseline | Unknown starting point | Sequence all isolates |
| Human PK unknown | Dosing uncertain | Use COVID-19 dosing initially |

---

## 8. Data Sources

| Tool | Query | Purpose |
|------|-------|---------|
| NCBI_Taxonomy | coronavirus | Classification |
| UniProt_search | Novel coronavirus polyprotein | Target proteins |
| NvidiaNIM_alphafold2 | Mpro, RdRp sequences | Structure prediction |
| NvidiaNIM_diffdock | Top 24 drugs | Docking screen |
| ChEMBL_search_drugs | coronavirus antivirals | Drug candidates |
| PubMed_search_articles | coronavirus drug resistance | Literature |

---

Example 2: Drug-Resistant Bacterial Infection

User Query

"Hospital outbreak of carbapenem-resistant Klebsiella pneumoniae. What treatment options exist?"

Expected Output (Key Sections)

# Outbreak Intelligence Report: Carbapenem-Resistant Klebsiella pneumoniae (CRKP)

**Generated**: 2026-02-04 | **Status**: URGENT - Active Outbreak

---

## Executive Summary

CRKP infections have limited treatment options. **Top candidates**:

1. **Ceftazidime-avibactam** ★★★ - BL/BLI combination (if KPC-producing)
2. **Meropenem-vaborbactam** ★★★ - BL/BLI combination
3. **Cefiderocol** ★★☆ - Siderophore cephalosporin (last resort)

**CRITICAL**: Carbapenemase type must be determined to guide therapy.

| Carbapenemase | Recommended Treatment |
|---------------|----------------------|
| **KPC** | Ceftazidime-avibactam OR meropenem-vaborbactam |
| **MBL (NDM, VIM)** | Cefiderocol OR aztreonam-avibactam |
| **OXA-48** | Ceftazidime-avibactam |

---

## 2. Resistance Mechanism Analysis

### 2.1 Carbapenemase Types

| Enzyme | Class | Prevalence | Key Feature |
|--------|-------|------------|-------------|
| KPC | Class A | 60-70% | Inhibited by avibactam |
| NDM | Class B (MBL) | 15-20% | NOT inhibited by avibactam |
| OXA-48 | Class D | 10-15% | Variable inhibition |
| VIM | Class B (MBL) | 5-10% | Requires cefiderocol |

### 2.2 Critical Testing

**MUST obtain before treatment**:
1. Carbapenemase gene testing (PCR or phenotypic)
2. Antimicrobial susceptibility testing
3. MIC for candidate agents

---

## 4. Treatment Options

### 4.1 Ranked by Evidence (KPC-Producing CRKP)

| Rank | Drug | Mechanism | Evidence | Notes |
|------|------|-----------|----------|-------|
| 1 | **Ceftazidime-avibactam** | BL + BLI | ★★★ RCT data | First-line for KPC |
| 2 | **Meropenem-vaborbactam** | BL + BLI | ★★★ RCT data | Alternative |
| 3 | **Imipenem-relebactam** | BL + BLI | ★★☆ Limited data | If others fail |
| 4 | Tigecycline | Protein synthesis | ★☆☆ High mortality | Last resort |
| 5 | Colistin | Cell membrane | ★☆☆ Nephrotoxicity | Last resort |

### 4.2 For MBL-Producing CRKP (NDM, VIM)

| Drug | Activity | Evidence |
|------|----------|----------|
| **Cefiderocol** | Active | ★★☆ RCT, mortality concern |
| **Aztreonam-avibactam** | Active | ★★☆ Phase 3 |
| Ceftazidime-avibactam | NOT active | ❌ MBL hydrolyzes |

---

## 6. Recommendations

### 6.1 Immediate Actions

1. **Test carbapenemase type** - Critical for drug selection
2. **Start empiric therapy** - Ceftazidime-avibactam + aminoglycoside
3. **Infection control** - Contact precautions, cohorting
4. **ID consult** - All CRKP cases

### 6.2 Empiric Therapy (Pending Testing)

If KPC suspected (most common): → Ceftazidime-avibactam 2.5g IV q8h

If MBL suspected (NDM endemic area): → Cefiderocol 2g IV q8h OR Aztreonam + avibactam

If critically ill: → Combination therapy until susceptibilities available

---

Example 3: Limited Information Scenario

User Query

"Novel virus isolated from agricultural workers with respiratory symptoms. Genome sequenced but not yet classified. What can we do?"

Expected Output (Key Sections)

# Outbreak Intelligence Report: Unclassified Respiratory Virus

**Status**: URGENT - Limited Data Available

---

## Executive Summary

With genome sequence available but taxonomic classification pending, we can still identify drug targets through sequence homology.

**Approach**:
1. BLAST genome against viral databases → Identify closest relatives
2. Predict key proteins → Prioritize by conservation
3. Dock broad-spectrum antivirals → Generate initial candidate list

**Initial candidates** (pending classification):
- Broad-spectrum polymerase inhibitors (favipiravir, ribavirin)
- Protease inhibitors (if protease identified)

---

## 1. Preliminary Classification

### 1.1 BLAST Results (Top Hits)

| Organism | Identity | E-value | Interpretation |
|----------|----------|---------|----------------|
| Influenza A H5N1 | 72% | 1e-89 | Possible orthomyxovirus |
| Influenza B | 65% | 1e-67 | Similar family |
| Thogoto virus | 58% | 1e-45 | Distant relative |

**Provisional Classification**: Likely orthomyxovirus (influenza-like)

### 1.2 Implications

If orthomyxovirus confirmed:
- **Neuraminidase inhibitors** (oseltamivir, zanamivir) may be active
- **Cap-dependent endonuclease inhibitors** (baloxavir) may be active
- **M2 inhibitors** likely inactive (resistance)

---

## 4. Preliminary Drug Screen

### 4.1 Broad-Spectrum Antivirals (No Classification Needed)

| Drug | Target | Activity Against | Priority |
|------|--------|------------------|----------|
| Favipiravir | RdRp | RNA viruses broadly | HIGH |
| Ribavirin | Multiple | RNA viruses broadly | MEDIUM |
| Remdesivir | RdRp | Coronaviruses mainly | LOW |

### 4.2 If Orthomyxovirus Confirmed

| Drug | Target | Activity | Priority |
|------|--------|----------|----------|
| **Oseltamivir** | Neuraminidase | Influenza A/B | HIGH |
| **Baloxavir** | Endonuclease | Influenza A/B | HIGH |
| Zanamivir | Neuraminidase | Influenza A/B | MEDIUM |

---

## 6. Recommendations

### With Limited Data

1. **Complete taxonomic classification** - Enables targeted drug selection
2. **Empiric broad-spectrum** - Favipiravir if RNA virus likely
3. **If influenza-like** - Prepare oseltamivir and baloxavir
4. **Supportive care** - Critical until specific therapy identified

### Research Priorities

1. Full genome annotation
2. In vitro antiviral testing
3. Animal model development
4. Serological assay development

Infectious Disease Outbreak Intelligence - Phase Details

Phase 1: Pathogen Identification

1.1 Taxonomic Classification

def identify_pathogen(tu, pathogen_query):
    """Classify pathogen taxonomically."""
    taxonomy = tu.tools.NCBIDatasets_suggest_taxonomy(query=pathogen_query)
    return {
        'taxid': taxonomy.get('taxid'),
        'scientific_name': taxonomy.get('scientific_name'),
        'rank': taxonomy.get('rank'),
        'lineage': taxonomy.get('lineage'),
        'type': classify_type(taxonomy)  # virus, bacteria, fungus, parasite
    }

1.2 Related Pathogens (Knowledge Transfer)

def find_related_pathogens(tu, taxid):
    """Find related pathogens for drug knowledge transfer."""
    relatives = tu.tools.NCBI_Taxonomy_get_children(taxid=taxid, rank="genus")
    related_with_drugs = []
    for rel in relatives:
        drugs = tu.tools.ChEMBL_search_targets(
            query=rel['scientific_name'], organism_contains=True
        )
        if drugs:
            related_with_drugs.append({'pathogen': rel, 'drugs': drugs})
    return related_with_drugs

1.3 Output Example

## 1. Pathogen Profile

### 1.1 Taxonomic Classification

| Property | Value |
|----------|-------|
| **Organism** | SARS-CoV-2 |
| **Taxonomy ID** | 2697049 |
| **Type** | RNA virus (positive-sense, single-stranded) |
| **Family** | Coronaviridae |
| **Genus** | Betacoronavirus |

### 1.2 Related Pathogens with Drug Precedent

| Relative | Similarity | Approved Drugs | Relevance |
|----------|------------|----------------|-----------|
| SARS-CoV | 79% genome | Remdesivir (EUA) | High |
| MERS-CoV | 50% genome | None approved | Medium |

---

Phase 2: Target Identification

2.1 Essential Protein Identification

def identify_targets(tu, pathogen_name):
    """Identify essential druggable targets."""
    proteins = tu.tools.UniProt_search(
        query=f"organism:{pathogen_name}", reviewed=True
    )
    targets = []
    for protein in proteins:
        chembl_target = tu.tools.ChEMBL_search_targets(query=protein['gene_name'])
        targets.append({
            'uniprot': protein['accession'],
            'name': protein['protein_name'],
            'function': protein['function'],
            'has_drug_precedent': len(chembl_target) > 0,
            'druggability': assess_druggability(protein)
        })
    return rank_targets(targets)

2.2 Target Prioritization Criteria

---

Phase 3: Structure Prediction

def predict_target_structure(tu, sequence, target_name):
    """Predict structure using AlphaFold2 via NVIDIA NIM."""
    structure = tu.tools.NvidiaNIM_alphafold2(
        sequence=sequence, algorithm="mmseqs2", relax_prediction=False
    )
    plddt_scores = parse_plddt(structure)
    return {
        'structure': structure['structure'],
        'mean_plddt': np.mean(plddt_scores),
        'high_confidence_regions': get_high_confidence(plddt_scores),
        'predicted_binding_site': identify_binding_site(structure)
    }

pLDDT Quality Assessment

---

Phase 4: Drug Repurposing Screen

4.1 Identify Candidates

def get_repurposing_candidates(tu, target_name, pathogen_family):
    """Find approved drugs to repurpose."""
    candidates = []
    # 1. Drugs approved for related pathogens
    candidates.extend(tu.tools.ChEMBL_search_drugs(
        query=pathogen_family, max_phase=4))
    # 2. Broad-spectrum antivirals
    candidates.extend(tu.tools.ChEMBL_search_drugs(
        query="broad spectrum antiviral", max_phase=4))
    # 3. Drugs with known activity against target class
    candidates.extend(tu.tools.DGIdb_get_drug_gene_interactions(
        genes=[target_name]))
    return deduplicate(candidates)

4.2 Docking Screen (NVIDIA NIM)

def dock_candidates(tu, target_structure, candidate_smiles_list):
    """Dock candidate drugs against target."""
    results = []
    for smiles in candidate_smiles_list:
        docking = tu.tools.NvidiaNIM_diffdock(
            protein=target_structure, ligand=smiles, num_poses=5
        )
        results.append({
            'smiles': smiles,
            'top_score': docking['poses'][0]['confidence'],
            'poses': docking['poses']
        })
    return sorted(results, key=lambda x: x['top_score'], reverse=True)

---

Phase 4.5: Pathway Analysis

def analyze_pathogen_pathways(tu, pathogen_name, pathogen_type):
    """Identify druggable metabolic pathways in pathogen."""
    pathways = tu.tools.kegg_search_pathway(query=f"{pathogen_name} metabolism")
    essential_genes = tu.tools.KEGG_get_pathway_genes(
        pathway_id=pathways[0]['pathway_id'])
    host_pathogen = tu.tools.kegg_search_pathway(
        query=f"{pathogen_name} host interaction")
    return {
        'metabolic_pathways': pathways,
        'essential_genes': essential_genes,
        'host_interaction': host_pathogen
    }

---

Phase 5: Literature Intelligence

def comprehensive_outbreak_literature(tu, pathogen_name):
    """Search all literature sources for outbreak intelligence."""
    pubmed = tu.tools.PubMed_search_articles(
        query=f"{pathogen_name} AND (outbreak OR treatment OR drug)",
        limit=50, sort="date")
    biorxiv = tu.tools.BioRxiv_list_recent_preprints(
        query=f"{pathogen_name} treatment mechanism", limit=20)
    medrxiv = tu.tools.MedRxiv_get_preprint(
        query=f"{pathogen_name} clinical trial", limit=20)
    arxiv = tu.tools.ArXiv_search_papers(
        query=f"{pathogen_name} drug discovery", category="q-bio", limit=10)
    trials = tu.tools.search_clinical_trials(
        condition=pathogen_name, status="Recruiting")

    key_papers = pubmed[:10]
    for paper in key_papers:
        citation = tu.tools.openalex_search_works(query=paper['title'], limit=1)
        paper['citations'] = citation[0].get('cited_by_count', 0) if citation else 0

    return {
        'pubmed': pubmed, 'biorxiv': biorxiv, 'medrxiv': medrxiv,
        'arxiv': arxiv, 'trials': trials, 'key_papers': key_papers
    }

Note: Preprints (BioRxiv/MedRxiv) are NOT peer-reviewed but CRITICAL for outbreak intelligence. Always note this caveat in reports.

Outbreak Intelligence Report: [PATHOGEN]

Generated: [Date] | Query: [Original query] | Status: In Progress

---

Executive Summary

[Analyzing...]

---

1. Pathogen Profile

1.1 Classification

[Analyzing...]

1.2 Related Pathogens

[Analyzing...]

---

2. Druggable Targets

2.1 Prioritized Targets

[Analyzing...]

2.2 Target Details

[Analyzing...]

---

3. Target Structures

3.1 Prediction Results

[Analyzing...]

3.2 Binding Sites

[Analyzing...]

---

4. Drug Repurposing Screen

4.1 Candidate Drugs

[Analyzing...]

4.2 Docking Results

[Analyzing...]

4.3 Top Candidates

[Analyzing...]

---

4.5 Pathway Analysis

Pathogen Metabolic Pathways (KEGG)

[Analyzing...]

Druggable Pathway Targets

[Analyzing...]

Host-Pathogen Interaction Points

[Analyzing...]

---

5. Literature Intelligence

5.1 Recent Findings

[Analyzing...]

5.2 Clinical Trials

[Analyzing...]

---

6. Recommendations

6.1 Immediate Actions

[Analyzing...]

6.2 Clinical Trial Opportunities

[Analyzing...]

6.3 Research Priorities

[Analyzing...]

---

7. Data Gaps & Limitations

[Analyzing...]

---

8. Data Sources

[Will be populated...]

---

Evidence Grading

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

Infectious Disease Outbreak Intelligence - Tool Reference

Phase 1: Pathogen Identification

NCBI Taxonomy Tools

Example - Classify pathogen:

# Search for pathogen
tax = tu.tools.NCBIDatasets_suggest_taxonomy(query="SARS-CoV-2")
# Returns: {"taxid": 2697049, "scientific_name": "...", "lineage": [...]}

UniProt Protein Tools

Example - Get viral proteins:

# Search for viral proteins
proteins = tu.tools.UniProt_search(
    query="organism:2697049",  # SARS-CoV-2 TaxID
    reviewed=True
)

---

Phase 2: Target Identification

ChEMBL Target Tools

Example - Find drug precedent:

# Search for protease inhibitors
drugs = tu.tools.ChEMBL_search_drugs(
    query="main protease coronavirus",
    max_phase=4  # Approved drugs only
)

DGIdb Tools

---

Phase 3: Structure Prediction (NVIDIA NIM)

Structure Prediction Tools

Example - Predict target structure:

# High-accuracy prediction
structure = tu.tools.NvidiaNIM_alphafold2(
    sequence=protease_sequence,
    algorithm="mmseqs2",
    relax_prediction=False
)
# Returns: {"structure": "<PDB content>", "plddt": [...]}

Structure Validation

def assess_structure_quality(structure_result):
    """Assess structure quality for docking."""
    plddt = structure_result.get('plddt', [])
    
    mean_plddt = np.mean(plddt)
    high_conf = sum(1 for p in plddt if p > 90) / len(plddt)
    
    return {
        'mean_plddt': mean_plddt,
        'high_confidence_fraction': high_conf,
        'docking_suitable': mean_plddt > 70 and high_conf > 0.5
    }

---

Phase 4: Drug Repurposing

ChEMBL Drug Search

DrugBank Tools

Docking Tools (NVIDIA NIM)

Example - Dock drug candidates:

# Dock drug against target
result = tu.tools.NvidiaNIM_diffdock(
    protein=target_pdb_content,
    ligand=drug_smiles,
    num_poses=10
)
# Returns: {"poses": [{"confidence": 0.94, "coordinates": ...}, ...]}

---

Phase 4.5: Pathway Analysis (NEW)

KEGG Pathway Tools

Example - Pathogen metabolism pathways:

# Search for viral replication pathways
pathways = tu.tools.kegg_search_pathway(
    query="coronavirus replication"
)

# Get essential genes
genes = tu.tools.KEGG_get_pathway_genes(
    pathway_id="ko03030"  # DNA replication
)

Reactome Tools

Example - Host-pathogen interaction pathways:

# Host response to infection
pathways = tu.tools.ReactomeContent_search(
    query="viral infection response",
    species="Homo sapiens"
)

---

Phase 5: Literature Intelligence (ENHANCED)

PubMed Tools

Example - Search outbreak literature:

papers = tu.tools.PubMed_search_articles(
    query="SARS-CoV-2 treatment drug",
    limit=50,
    sort="date"
)

Preprint Servers (CRITICAL for Outbreaks)

⚠️ Preprints are NOT peer-reviewed but critical for emerging outbreaks!

Example - Search preprints (bioRxiv/medRxiv don't have search APIs, use EuropePMC):

# Search for newest preprint findings
preprints = tu.tools.EuropePMC_search_articles(
    query=f"{pathogen_name} mechanism resistance",
    source="PPR",  # PPR = Preprints (bioRxiv, medRxiv, etc.)
    pageSize=20
)

# If you have a specific DOI, retrieve full metadata:
if doi_from_search.startswith('10.1101/'):
    full_preprint = tu.tools.BioRxiv_get_preprint(doi=doi_from_search)

# Alternative: Use web search for bioRxiv
web_results = tu.tools.web_search(
    query=f"{pathogen_name} clinical trial effectiveness",
    limit=20
)

# Computational papers
arxiv = tu.tools.ArXiv_search_papers(
    query=f"{pathogen_name} drug discovery",
    category="q-bio",
    limit=10
)

Citation Analysis Tools

Example - Find high-impact papers:

# Get citation counts
papers = tu.tools.openalex_search_works(
    query="remdesivir COVID-19 trial",
    limit=20
)
# Returns: {"cited_by_count": 5234, ...}

# AI-ranked papers
ranked = tu.tools.SemanticScholar_search_papers(
    query="SARS-CoV-2 drug resistance",
    limit=20
)

Clinical Trials Tools

Example - Find active trials:

trials = tu.tools.search_clinical_trials(
    condition="COVID-19",
    intervention="antiviral",
    status="Recruiting"
)

---

Workflow Code Examples

Example 1: Complete Outbreak Analysis

def analyze_outbreak(tu, pathogen_name):
    """Complete outbreak intelligence workflow."""
    
    # Phase 1: Identify pathogen
    taxonomy = tu.tools.NCBIDatasets_suggest_taxonomy(query=pathogen_name)
    taxid = taxonomy['taxid']
    
    # Phase 2: Get target proteins
    proteins = tu.tools.UniProt_search(
        query=f"organism:{taxid}",
        reviewed=True
    )
    
    # Phase 3: Predict structures for top targets
    structures = {}
    for protein in proteins[:3]:  # Top 3 targets
        seq = tu.tools.UniProt_get_sequence_by_accession(
            accession=protein['accession']
        )
        struct = tu.tools.NvidiaNIM_alphafold2(sequence=seq)
        structures[protein['name']] = struct
    
    # Phase 4: Find repurposing candidates
    candidates = tu.tools.ChEMBL_search_drugs(
        query=f"{pathogen_name} OR broad spectrum antiviral",
        max_phase=4
    )
    
    # Dock top candidates
    docking_results = []
    for drug in candidates[:20]:
        result = tu.tools.NvidiaNIM_diffdock(
            protein=structures['main_protease']['structure'],
            ligand=drug['smiles'],
            num_poses=5
        )
        docking_results.append({
            'drug': drug,
            'score': result['poses'][0]['confidence']
        })
    
    # Phase 5: Literature search
    papers = tu.tools.PubMed_search_articles(
        query=f"{pathogen_name} treatment",
        limit=50
    )
    
    return {
        'taxonomy': taxonomy,
        'targets': proteins,
        'structures': structures,
        'drug_candidates': sorted(docking_results, 
                                  key=lambda x: x['score'], 
                                  reverse=True),
        'literature': papers
    }

Example 2: Rapid Drug Screen

def rapid_drug_screen(tu, target_sequence, drug_smiles_list):
    """Rapid docking screen for drug repurposing."""
    
    # Quick structure prediction
    structure = tu.tools.NvidiaNIM_esmfold(sequence=target_sequence)
    
    # Dock all candidates
    results = []
    for smiles in drug_smiles_list:
        docking = tu.tools.NvidiaNIM_diffdock(
            protein=structure['structure'],
            ligand=smiles,
            num_poses=3
        )
        results.append({
            'smiles': smiles,
            'score': docking['poses'][0]['confidence']
        })
    
    return sorted(results, key=lambda x: x['score'], reverse=True)

Example 3: Knowledge Transfer from Related Pathogen

def transfer_knowledge(tu, novel_pathogen, reference_pathogen):
    """Transfer drug knowledge from related pathogen."""
    
    # Get drugs approved for reference pathogen
    ref_drugs = tu.tools.ChEMBL_search_drugs(
        query=reference_pathogen,
        max_phase=4
    )
    
    # Get target from novel pathogen
    novel_proteins = tu.tools.UniProt_search(
        query=f"organism:{novel_pathogen}"
    )
    
    # Find homologous targets
    homologs = []
    for protein in novel_proteins:
        # BLAST against reference
        blast = tu.tools.BLAST_protein_search(
            sequence=protein['sequence'],
            database="refseq_protein",
            organism=reference_pathogen
        )
        if blast and blast[0]['identity'] > 70:
            homologs.append({
                'novel_target': protein,
                'reference_homolog': blast[0],
                'identity': blast[0]['identity']
            })
    
    # Match drugs to homologous targets
    candidates = []
    for drug in ref_drugs:
        for homolog in homologs:
            if drug['target'] == homolog['reference_homolog']['accession']:
                candidates.append({
                    'drug': drug,
                    'target_homology': homolog['identity'],
                    'expected_activity': 'High' if homolog['identity'] > 90 else 'Medium'
                })
    
    return candidates

---

Fallback Chains

Taxonomy

Structure Prediction

Docking

Drug Search

Pathway Analysis (NEW)

Literature (ENHANCED)

---

Common Parameter Mistakes

---

NVIDIA NIM Requirements

API Key: NVIDIA_API_KEY environment variable required

Rate limits: 40 RPM (1.5 second minimum between calls)

Async operations:

• AlphaFold2 may return 202, requiring polling
• ESMFold is synchronous (faster for rapid screening)

Check Availability

import os

nvidia_available = bool(os.environ.get("NVIDIA_API_KEY"))
if not nvidia_available:
    print("Warning: NVIDIA NIM tools unavailable, using fallbacks")

---

Speed Optimization

For Urgent Outbreaks

1. Use ESMFold first for rapid structure (30 sec vs 5-15 min) 2. Dock FDA-approved only initially (fastest to deploy) 3. Parallelize docking if possible 4. Cache structures for repeated queries

Prioritization Order

def prioritize_candidates(candidates):
    """Prioritize by speed to clinical use."""
    return sorted(candidates, key=lambda x: (
        -x['fda_approved'],      # FDA approved first
        -x['phase'],             # Higher phase next
        -x['docking_score']      # Then by score
    ))