Validation Best Practices

Last Updated: November 29, 2025

Computational predictions in virome analysis require validation. This guide covers strategies for validating viral identifications, host predictions, and functional annotations.

Why Validation Matters

Computational predictions have limitations: - Viral identification: 10-30% false positives - Host prediction: 30-60% accuracy for novel phages - Functional annotation: 40-70% hypothetical proteins

Validation provides: - Confidence in results - Publishable claims - Biological insights - Method benchmarking

Viral Identification Validation

In Silico Validation

1. CheckV Quality Assessment

# Essential first step
checkv end_to_end viral_contigs.fa checkv_out/ -t 8

# Red flags (likely false positives):
# - High contamination (>10%)
# - Very low completeness (<10%)
# - No viral genes identified
# - Contamination warning in provirus column

# Filter false positives
awk -F'\t' '$8 != "Not-determined" && $10 < 5 {print $1}' \
    checkv_out/quality_summary.tsv > validated_ids.txt

2. Gene Content Analysis

# Check for viral hallmark genes
# Expected in most DNA viruses:
# - Terminase, portal protein, major capsid protein

# Search for hallmark genes
hmmsearch --tblout hallmark_hits.txt viral_hallmark_hmms.hmm proteins.faa

# Contigs with ≥1 hallmark gene = higher confidence

3. Manual Inspection

# For key findings, manually inspect:
# - Gene synteny (organization)
# - Protein homology
# - Genomic context

# View in Artemis or GenBank
artemis contig.gbk

Experimental Validation

1. PCR Confirmation

# Design primers flanking predicted viral region
# Positive PCR = virus present in sample

# qPCR for quantification
qpcr_primers viral_gene.fa

# Expected: Amplicon if virus is real, no amplicon if false positive

2. Plaque Assays (Cultivable Phages)

# Gold standard for phages
# 1. Isolate phage from environmental sample
# 2. Purify plaques
# 3. Sequence isolate
# 4. Compare to predicted sequence

# Match? → Validated!
# No match? → Prediction may be wrong or virus is uncultivable

3. Viral Metagenomics Comparison

# Compare VLP-enriched vs total metagenome
# True viruses should be:
# - Enriched in VLP fraction (10-100x)
# - Depleted in non-VLP fraction

# Calculate enrichment
coverage_VLP / coverage_total > 10  # Likely viral

Host Prediction Validation

Computational Validation

1. Taxonomic Consistency

# Check if predictions are taxonomically reasonable
# E.g., marine phage predicted to infect human gut bacteria = suspicious

def check_taxonomic_consistency(phage_env, host_env):
    """Check if phage and host environments match"""
    compatible = {
        'marine': ['marine', 'ocean', 'seawater'],
        'gut': ['gut', 'fecal', 'intestinal'],
        'soil': ['soil', 'terrestrial', 'sediment']
    }

    for env_type, keywords in compatible.items():
        if any(k in phage_env for k in keywords):
            if any(k in host_env for k in keywords):
                return True
    return False

2. Prophage Analysis

# If predicted host genome available, check for integrated prophages

# Search for prophage in host genome
phispy predicted_host_genome.fa -o prophage_check/

# If phage sequence found integrated → Strong validation!

3. CRISPR Validation

# Gold standard: CRISPR spacer match
# Indicates phage-host encounter

blast spacer vs phage_genome
# >95% identity, <2 mismatches = high confidence

Experimental Validation

1. Infection Assays

# Culture-based (if phage and host are cultivable)
# 1. Grow predicted host
# 2. Add phage lysate
# 3. Monitor for lysis or plaques

# Lysis/plaques → Validated host!
# No lysis → Either wrong prediction or resistance

2. qPCR Co-occurrence

# Quantify phage and predicted host across samples
# Strong positive correlation suggests interaction

# Calculate Spearman correlation
correlation(phage_abundance, host_abundance)
# ρ > 0.6, p < 0.05 → Supporting evidence

3. Hi-C Proximity Ligation

# Detects physical proximity of phage-host DNA
# Indicates active infection

# Hi-C reads linking phage to host = strong validation
# Requires specialized sequencing

4. Single-Cell Genomics

# Sequence single infected cells
# Phage + host DNA in same cell = validation

# Use flow cytometry to sort infected cells
# Sequence with MDA or similar

Functional Annotation Validation

In Silico Validation

1. Domain Architecture

# Check if protein domains are consistent with function
# E.g., "DNA polymerase" should have polymerase domains

# Run InterProScan
interproscan.sh -i protein.faa -f tsv -o domains.tsv

# Check consistency
# Predicted: Terminase
# Domains: Terminase_ATPase, Terminase_nuclease
# → Consistent, validated!

2. Phylogenetic Placement

# Place protein in phylogenetic tree with known proteins
# Should cluster with proteins of similar function

# Build tree
mafft --auto protein_with_refs.faa > aligned.faa
iqtree -s aligned.faa -m TEST -bb 1000

# Check: Does it cluster with annotated terminases?
# Yes → Validated
# No → Annotation may be wrong

3. Structure Prediction

# Use AlphaFold to predict structure
# Compare to known structures

alphafold --fasta protein.faa --output alphafold_out/

# Compare to PDB
# Similar structure to known terminase? → Validated

Experimental Validation

1. Heterologous Expression

# Clone gene into expression vector
# Express in E. coli
# Test predicted activity

# E.g., if predicted as "endolysin":
# - Express protein
# - Test lytic activity on bacterial cells
# - Activity → Validated!

2. Deletion/Mutation Studies

# Delete or mutate predicted essential gene
# Phenotype should match prediction

# E.g., delete predicted "portal protein"
# Expected: Non-functional phage
# Observed: Non-functional phage → Validated!

Abundance Estimate Validation

Technical Replicates

# Sequence same sample multiple times
# Abundance should be highly correlated

# Expected correlation: r > 0.95
# Lower correlation indicates technical noise

Spike-In Controls

# Add known amount of control virus to sample
# Quantify after sequencing

observed_abundance / expected_abundance ≈ 1
# Deviations indicate bias or error

qPCR Validation

# Absolute quantification via qPCR
# Compare to sequencing-based abundance

# Should be correlated (not necessarily 1:1)
# Correlation r > 0.7 = reasonable agreement

Validation Confidence Levels

Evidence Type	Confidence Level	Suitable For
CRISPR spacer match	Very High (90-95%)	Host prediction
Plaque assay + sequencing	Very High (95%+)	Viral ID, host
Hi-C proximity	High (80-90%)	Host prediction
CheckV Complete + hallmark genes	High (85-90%)	Viral ID
qPCR confirmation	Medium-High (75-85%)	Abundance, presence
Taxonomic consistency	Medium (60-70%)	Host prediction
Multiple tools agree	Medium (65-75%)	Viral ID
Single prediction method	Low (30-50%)	Exploratory only

When to Validate

Always validate: - Novel virus claims (new species/family) - Key biological findings (e.g., "this phage controls bloom") - Host predictions used for downstream analysis - Functional claims about gene products - Unexpected or controversial results

Can skip validation: - Exploratory/pilot studies (but acknowledge limitation) - When using multiple high-quality methods (e.g., CRISPR + homology) - Well-established results (e.g., T4-like phage identified)

Validation Checklist

Before submitting for publication:

[ ] Viral predictions validated with CheckV + manual inspection
[ ] Key viruses checked for hallmark genes
[ ] Negative controls show <1% of sample reads
[ ] Host predictions have ≥2 lines of evidence for key findings
[ ] Functional annotations consistent with domain architecture
[ ] Abundance estimates correlated with technical replicates (if available)
[ ] Unexpected findings validated experimentally or noted as predictions
[ ] Limitations acknowledged in text

Reporting Validation

Good example:

"We identified 234 viral contigs using a consensus of VirSorter2, VIBRANT, and geNomad (≥2 tools agreeing). Quality assessment with CheckV identified 89 high-quality viral genomes (>50% complete, <5% contamination). We validated the presence of 10 randomly selected viruses via PCR (9/10 positive, Supplementary Fig. 3). Host predictions were based on CRISPR spacer matches (n=12, high confidence) or iPHoP consensus predictions with >90 confidence score (n=45, medium confidence)."

Bad example:

"We found 234 viruses. Host predictions were made with iPHoP." (No validation, no quality assessment, no confidence levels)

Validation Best Practices

Why Validation Matters

Viral Identification Validation

In Silico Validation

Experimental Validation

Host Prediction Validation

Computational Validation

Experimental Validation

Functional Annotation Validation

In Silico Validation

Experimental Validation

Abundance Estimate Validation

Technical Replicates

Spike-In Controls

qPCR Validation

Validation Confidence Levels

When to Validate

Validation Checklist

Reporting Validation

Further Reading