| name | phage-genomics-analyst |
| description | Phage genomics specialist for bacteriophage genome analysis, taxonomy classification, lifecycle prediction, anti-CRISPR detection, and PhageScope dataset workflows. Use for phage diversity studies, host prediction, comparative genomics, and phylogenetic analysis of bacteriophages. |
Phage Genomics Analyst
Expert-level skill for bacteriophage genome analysis, covering taxonomy, lifecycle prediction, anti-CRISPR systems, and PhageScope dataset workflows.
Core Capabilities
1. Phage Taxonomy Classification
ICTV Classification System (2023 - Major Reorganization)
⚠️ IMPORTANT: ICTV 2022-2023 reorganized Caudovirales into 20+ families based on phylogenomics.
The traditional morphology-based families (Myoviridae, Siphoviridae, Podoviridae) have been dissolved.
Order Caudovirales (tailed phages, ~96% of known phages):
New phylogenomics-based families (examples):
-
Herelleviridae: T4-like phages (contractile tails)
- Genome: 160-170 kb, GC 35-40%
- Key markers: gp18 (tail sheath), gp19 (tail tube), gp23 (major capsid)
- Examples: T4, P1, Mu
-
Drexlerviridae: λ-like phages (long non-contractile tails)
- Genome: 40-60 kb, GC 45-55%
- Key markers: Integrase (int), CI repressor, tail length protein
- Examples: λ, HK97, N15
-
Rountreeviridae: T7-like phages (short tails)
- Genome: 40-75 kb, GC 48-52%
- Key markers: RNA polymerase, DNA polymerase, tail spike proteins
- Examples: T7, P22, Φ29
Legacy morphology terms (still used informally in literature):
- Myoviridae morphology: Contractile tails (now split into Herelleviridae, Peduoviridae, etc.)
- Siphoviridae morphology: Long non-contractile tails (now split into Drexlerviridae, Siphoviridae sensu stricto, etc.)
- Podoviridae morphology: Short non-contractile tails (now split into Rountreeviridae, Autographiviridae, etc.)
Note: When analyzing phage genomes, use modern ICTV 2023 family names. Legacy terms are acceptable for morphological descriptions but not for taxonomic classification.
Other Orders:
- Tubulavirales: Filamentous phages (M13, fd)
- Levivirales: ssRNA phages (MS2, Qβ)
- Corticovirales: Lipid-containing phages (PRD1)
Taxonomic Assignment Workflow
-
BLAST-based: Use terminase large subunit (TerL) as marker
- Identity >70%: Same genus
- Identity 40-70%: Same family
- Identity <40%: Different family
-
vConTACT2: Network-based clustering
- Uses gene-sharing networks
- More accurate for novel phages
-
CheckV: Quality assessment
- Completeness: >90% = high-quality
- Contamination: <5% = clean
2. Lifecycle Prediction (Lytic vs Temperate)
Temperate Phage Markers
Integration System:
- Integrase (int): Tyrosine or serine recombinase
- Tyrosine integrase: λ-like, att site recombination
- Serine integrase: ΦC31-like, directional recombination
- Excisionase (xis): Often co-located with int
- attP site: Phage attachment site (15-50 bp core)
Lysogeny Maintenance:
- CI repressor: Helix-turn-helix DNA binding
- Auto-regulation of PR and PRM promoters
- Cleavage by RecA* during SOS response
- CII/CIII: Lysogeny establishment (λ-like)
- Anti-repressor: Counteracts CI in some phages
Lytic Phage Markers
Lysis System:
- Endolysin: Peptidoglycan hydrolase
- SAR endolysin: Secretion-accumulation-release
- Holin-endolysin: Holin creates holes
- Holin: Small membrane protein
- Class I: 3 TMDs (λ S protein)
- Class II: 2 TMDs (P22 13 protein)
- Class III: 1 TMD (Φ29 gp14)
- Spanin: Outer membrane disruption
- i-spanin: Inner membrane anchored
- o-spanin: Outer membrane anchored
- Unimolecular spanin: Single protein (Φ29 gp15)
No Integration:
- Absence of integrase/excisionase
- No CI-like repressor
- Direct lytic cycle
Prediction Workflow
1. Search for integrase (HMM: PF00589, PF02899)
- Found → Temperate candidate
- Not found → Check for CI repressor
2. Search for CI repressor (HMM: PF00126, PF01381)
- PF00126: HTH_1 (general helix-turn-helix domain)
- PF01381: HTH_11 (phage repressor-specific HTH)
- Found → Temperate (even without int)
- Not found → Likely lytic
3. Verify with lysis genes
- Holin + Endolysin + Spanin → Complete lysis system
- SAR endolysin only → Alternative lysis
4. Check for anti-CRISPR
- Acr genes → Often temperate (prophage defense)
3. Anti-CRISPR Protein Analysis
Anti-CRISPR Families
Type I Systems:
- AcrIF1-10: Inhibit Type I-F (Pseudomonas)
- AcrIE1-9: Inhibit Type I-E (E. coli)
- Mechanism: Block Cas complex DNA binding or cleavage
Type II Systems:
- AcrIIA1-17: Inhibit Type II-A (Listeria, Streptococcus)
- AcrIIC1-5: Inhibit Type II-C (Neisseria)
- Mechanism: Mimic DNA, block Cas9 RuvC/HNH domains
Type V Systems:
- AcrVA1-5: Inhibit Type V-A (Cas12a)
- Mechanism: Acetylate crRNA, block DNA binding
Detection Workflow
-
Database Search:
- Anti-CRISPRdb (2023): 54 families, 300+ proteins
- HMM profiles for known Acr families
-
Genomic Context:
- Often near HTH transcriptional regulators
- Located in "anti-CRISPR associated" (Aca) regions
- Prophage regions (PHASTER/PhiSpy predictions)
-
Functional Validation:
- Co-occurrence with CRISPR-Cas in host
- Expression during lysogeny
- Inhibition assays (plaque reduction)
4. PhageScope Dataset Guide
Dataset Overview (2024 Release)
- Total sequences: 873,718 phage genomes
- Quality distribution:
- High-quality (>90% complete): ~45%
- Medium-quality (50-90%): ~35%
- Low-quality (<50%): ~20%
- Taxonomic coverage:
- Caudovirales: 96% (Myo: 35%, Sipho: 45%, Podoviridae: 16%)
- Other orders: 4%
Host Distribution
- Top hosts: E. coli (28%), Pseudomonas (15%), Streptococcus (12%)
- Host prediction confidence:
- Genus-level: 85% accuracy
- Species-level: 65% accuracy
- Recommended filtering:
- Use only high-quality genomes for ML training
- Balance host classes (min 100 genomes per genus)
- Remove duplicates (>95% ANI)
Key Metadata Fields
phage_id: Unique identifier
genome_length: bp
gc_content: %
num_genes: Predicted ORFs
host_genus: Predicted host genus
host_species: Predicted host species
completeness: CheckV completeness %
contamination: CheckV contamination %
lifestyle: lytic/temperate/unknown
family: ICTV family assignment
genus: ICTV genus assignment
5. Standard Analysis Workflows
Phage Genome Annotation Pipeline
1. Quality Control
- CheckV: completeness, contamination
- Filter: >90% complete, <5% contamination
2. Gene Prediction
- Prokka: Standard annotation
- PHANOTATE: Phage-specific (better for small genes)
- Combine: Use Prokka + PHANOTATE consensus
3. Functional Annotation
- BLASTp: nr database (e-value < 1e-5)
- HMMER: Pfam, TIGRFAM, custom phage HMMs
- InterProScan: Integrated domain search
4. Specialized Annotations
- tRNAs: tRNAscan-SE
- tmRNAs: ARAGORN
- CRISPR arrays: CRISPRCasFinder
- Anti-CRISPR: AcrDB search
- Virulence factors: VFDB
- AMR genes: CARD/ResFinder
5. Genome Visualization
- Clinker: Synteny plots
- PhageScope: Phage genome browser and annotation viewer
- DNAplotter: Circular genome maps
Diversity Analysis Workflow
1. Data Preparation
- Filter: High-quality genomes only
- Deduplicate: CD-HIT-EST (95% sequence identity) or FastANI (95% ANI)
- Subset: Max 500 genomes per host genus
2. Feature Extraction
- Genomic: Length, GC%, gene count, coding density
- Functional: tRNA count, AMR genes, virulence factors
- Structural: Capsid size (TerL phylogeny), tail type
3. Diversity Metrics
- Shannon entropy: H' = -Σ(pi × ln(pi))
- pi = proportion of genomes in category i
- Categories: Host genus, family, lifestyle
- Simpson index: D = 1 - Σ(pi²)
- Pielou evenness: J = H' / ln(S)
- S = number of categories
- Beta diversity: Bray-Curtis between host groups
4. Statistical Testing
- ANOVA/Kruskal-Wallis: Compare metrics across groups
- PERMANOVA: Beta diversity significance
- Post-hoc: Tukey HSD or Dunn's test
5. Visualization
- Bar plots: Diversity indices by group
- Heatmaps: Pairwise beta diversity
- PCoA: Ordination of beta diversity
Host Prediction ML Pipeline
1. Feature Engineering
- k-mer frequencies (k=4, 5, 6)
- Genomic signatures (GC skew, codon usage)
- Protein domain counts (Pfam)
- CRISPR spacer matches
2. Data Splitting
- Stratified by host genus
- Train: 70%, Val: 15%, Test: 15%
- Ensure no genome leakage (same phage in train/test)
3. Model Training
- Random Forest: Baseline (1000 trees)
- XGBoost: Gradient boosting (tune max_depth, learning_rate)
- Neural network: Optional (if >10k genomes)
4. Evaluation
- Metrics: Accuracy, F1-macro, AUC-ROC
- Cross-validation: 5-fold stratified
- Feature importance: SHAP values
5. Interpretation
- Top features: k-mers, domains, genomic signatures
- Misclassification analysis: Why certain hosts confused?
- Confidence thresholds: Only predict if P > 0.8
Comparative Genomics Workflow
1. Genome Selection
- Representative genomes per group (max 50)
- High-quality, complete genomes
- Balanced sampling across groups
2. Whole Genome Alignment
- Mauve: Multiple genome alignment
- progressiveMauve: Handle rearrangements
- Output: XMFA format
3. Synteny Analysis
- Clinker: Synteny visualization
- Identify: Conserved gene clusters
- Detect: Rearrangements, inversions
4. Pangenome Analysis
- Roary: Gene presence/absence
- Panaroo: Handle assembly errors
- Output: Core (95%), soft-core (80%), shell (20%), cloud (<20%)
5. Phylogenetic Analysis
- Marker genes: TerL, major capsid, portal protein
- Alignment: MAFFT (--auto)
- Tree: IQ-TREE (ModelFinder + ultrafast bootstrap)
- Visualization: ggtree (R) or iTOL
6. Recombination Detection
- PhiPack: PHI test for recombination
- Gubbins: Identify recombination regions
- ClonalFrameML: Correct for recombination in tree
Key Metrics Reference
Genome Quality Tiers
| Metric | High-Quality | Medium | Low |
|---|
| Completeness | >90% | 50-90% | <50% |
| Contamination | <5% | 5-10% | >10% |
| N50 | Full genome | >50 kb | <50 kb |
| tRNAs present | Yes | Partial | No |
Taxonomic Assignment Confidence
| Method | Genus-Level | Species-Level |
|---|
| TerL BLAST (>70%) | High | Medium |
| vConTACT2 (cluster) | High | Low |
| ANI (>95%) | High | High |
| Gene content (>80%) | Medium | Low |
Lifecycle Prediction Accuracy
| Marker | Sensitivity | Specificity |
|---|
| Integrase present | 85% | 95% |
| CI repressor | 75% | 90% |
| Both int + CI | 95% | 98% |
| Neither | 90% (lytic) | 85% |
Diversity Index Interpretation
| Index | Range | Interpretation |
|---|
| Shannon H' | 0-4+ | >2.5 = high diversity |
| Simpson 1-D | 0-1 | >0.7 = high diversity |
| Pielou J | 0-1 | >0.8 = even distribution |
| Bray-Curtis | 0-1 | <0.3 = similar communities |
Common Pitfalls
1. Misclassifying Temperate Phages
Problem: Calling temperate phages as lytic
Cause: Integrase too divergent for BLAST
Solution: Use HMM search (PF00589, PF02899) with e-value < 1e-3
2. Over-splitting Host Classes
Problem: Too many host species with <50 genomes each
Cause: Using species-level instead of genus-level
Solution: Aggregate to genus-level, filter min 100 genomes
3. Ignoring Quality Filters
Problem: Including low-quality genomes in analysis
Cause: Not running CheckV or using loose thresholds
Solution: Always filter >90% complete, <5% contamination
4. Data Leakage in ML
Problem: Same phage in train and test sets
Cause: Random splitting without deduplication
Solution: CD-HIT-EST (95% sequence identity) or FastANI (95% ANI) before splitting
5. Incorrect Phylogenetic Markers
Problem: Using 16S rRNA (not in phages)
Cause: Confusing bacterial and phage markers
Solution: Use TerL, major capsid, or portal protein
Scientific Communication Guidelines
When reporting phage genomics results:
- Genome Quality: Always report CheckV completeness/contamination
- Taxonomic Assignment: State method and confidence threshold
- Lifecycle Prediction: List evidence (int, CI, lysis genes)
- Host Prediction: Report confidence and method limitations
- Diversity Metrics: Include confidence intervals (bootstrap)
- Comparative Analysis: Specify alignment parameters and tree method
Additional Resources
For detailed references: