Crassify: Protein-Based Viral Taxonomy Tool

Crassify is a high-throughput, fast tool for computing relatedness between viral genomes using whole-proteome pairwise protein alignments. Designed for metagenomic datasets, it enables:

• Rapid viral species detection
• Novelty detection
• MAG completeness estimation
• Phylogenetic distance estimation

🧩 Where does Crassify fit in your pipeline?

Crassify can be used after viral contigs have been identified in your metagenomic data but can also be used as a quick method to identify viral contigs in metagenomes.

If you already ran tools such as VIBRANT or VirSorter2 - or any other viral discovery pipeline - you’ll often end up with a large set of putative viral genomes/contigs.

The next big question is: what are they, and how do they relate to known viruses?

That’s where Crassify comes in:

• Uses a curated database of ICTV-classified viral genomes (14,000+ references).
• Assigns taxonomy to your contigs by comparing entire proteomes, not just marker genes.
• Provides metrics like viral content, genome completeness, and novelty score to help decide if your contig represents a known virus or something new.
• Produces distances and summary visualizations for downstream phylogenetic or ecological analyses.

In short: Run your favorite viral discovery tool → feed the predicted viral contigs into Crassify → get taxonomy + relatedness to ICTV reference genomes.

Installation

Clone the repo and install dependencies:

git clone https://github.com/linda5mith/crassify.git
cd crassify/
mamba env create -f environment.yml
mamba activate crassify
# Install Crassify as a CLI command:
pip install -e .

Test installation

crassify -i sample_data/test_phages_nucl/pooled_test_phages.fna -o ~/crassify_test

Output

Crassify produces several output files:

• percentage_viral.csv — per-genome summary of viral content, completeness, and novelty.
• distances.csv — pairwise inter-genome distances and similarity metrics.
• crassify_summary.png — quick visualization of input genomes and viral content.

---

percentage_viral.csv

Column	Description
genome_ID	Accession or identifier of the query genome/contig
genome_length	Total nucleotide length of the query genome
protein_hits	Number of proteins with at least one significant match
top_species_hit	Best-matching reference virus species
top_species_hit_genome_accn	Accession of the top reference genome
sseqid_genome_length	Length of the best-matching reference genome
total_aln_length	Summed alignment length across all proteins
% contig viral	Fraction of query genome aligning to viral proteins
% contig completeness	Completeness relative to the top reference genome
#_proteins	Number of predicted proteins in the query genome
% proteins aligned	Percentage of proteins with hits in the reference DB
novelty_score	Higher = more novel (penalizes low alignment/completeness)
is_novel	Boolean flag (True if novelty_score > 60)

---

distances.csv

Column	Description
qseqid_genome_ID	Query genome accession/ID
sseqid_genome_ID	Reference genome accession/ID
sseqid_virus	Virus name of the reference genome
distance	Inter-genome distance (lower = more similar)
total_aln_length	Summed amino acid alignment length
avg_pid	Average % identity across alignments
avg_genome_length	Average length of query and reference genomes
qseqid_genome_length	Query genome length
sseqid_genome_length	Reference genome length

---

Example output visualization:

---

Input Files

Crassify takes either:

• Nucleotide sequences (that get translated), or
• Protein FASTA files (.faa)
  (All proteins for a given genome should be supplied together)

---

Building and Compiling Your Own Reference Database

You can build a custom Crassify-compatible database using your own set of viral proteomes.

Step 1: Create a DIAMOND Database

diamond makedb --in your_viral_proteomes.faa -d VIRAL_DB.dmnd

Step 2: Update/or add metadata corresponding to your viral genomes

Add metadata corresponding to your viral genomes in the format as seen in data/crassify_metadata.csv