VICAST: An Integrated Toolkit for Viral Genome Annotation Curation and Low-Frequency Variant Analysis in Passage Studies

VICAST is an integrated software toolkit designed to streamline viral passage studies by combining semi-automated genome annotation with manual curation and optimized low-frequency variant calling, offering superior speed and accuracy compared to existing tools for diverse viral families.

The Gist

Imagine you are a detective trying to solve a mystery about how a virus changes over time. You have a stack of old, messy maps (the virus's genetic code) and a pile of blurry photos (the DNA sequencing data). Your goal is to find the tiny, subtle changes in the virus that happen as it adapts to new environments, like a cell culture lab.

The problem? Most existing tools are like detectives who only look at the "final draft" of the story. They miss the tiny, early edits that happen before the story is finished. Also, many of the maps they use are incomplete or have the wrong street names.

Enter VICAST. Think of VICAST as a super-smart, high-tech detective's toolkit designed specifically for this job. It doesn't just find the changes; it fixes the maps first, checks for fake clues, and then tells you exactly what those changes mean.

Here is how VICAST works, broken down into simple analogies:

1. Fixing the Map (Annotation Curation)

Imagine the virus's genome is a giant instruction manual written in a foreign language.

• The Problem: Sometimes, the manual says "Make a big machine" (a polyprotein), but it doesn't tell you that this machine is actually made of 16 smaller, specific parts (like a screwdriver, a hammer, and a wrench). If you don't know which part broke, you can't fix it.
• The VICAST Solution: VICAST has a special "Map Maker" mode. It takes the messy manual and breaks it down into the specific parts.
  • The "Human Touch": Unlike other tools that just guess, VICAST pauses and says, "Hey, human expert, look at this part of the map. Does this look right?" It forces a human to double-check the work before moving on. This ensures the map is perfect.
  • The Result: Now, instead of saying "The big machine is broken," VICAST can say, "The screwdriver part of the machine is broken." This is huge because it tells scientists exactly which part of the virus is changing.

2. Finding the Tiny Clues (Low-Frequency Variant Analysis)

In a virus population, not every virus is identical. It's like a crowd of people where 95% are wearing red shirts, but 5% are wearing blue shirts.

• The Problem: Most tools only look at the "majority vote." They see the red shirts and ignore the blue ones. But in a virus lab, those "blue shirts" (the minority) are often the ones learning how to survive and become the next dominant strain.
• The VICAST Solution: VICAST is like a magnifying glass that can spot the 5% wearing blue shirts. It uses sensitive math to find these rare changes (even if they are only 3% of the crowd) and tells you exactly where they are on the map.

3. Checking for Fake Clues (Contamination Screening)

When you grow viruses in a lab, sometimes other things sneak in, like bacteria or fungi. It's like trying to listen to a specific singer in a room, but someone else is humming in the background.

• The Problem: If you don't notice the background noise, you might think the singer changed their tune, when actually, it was just the other person.
• The VICAST Solution: Before VICAST starts analyzing the virus, it runs a "sniffer test." It looks at the data and says, "Wait a minute, I smell E. coli or yeast here!" It flags these intruders so scientists don't waste time analyzing fake clues.

4. Connecting the Dots (Haplotype Reconstruction)

Sometimes, a virus has two changes: one on its left arm and one on its right leg. Are these changes happening on the same virus, or are they on two different viruses?

• The Problem: Standard tools see the changes but don't know if they are linked.
• The VICAST Solution: VICAST looks at the original "photos" (the DNA reads) to see if the two changes appear together on the same piece of paper. It's like checking if a person is wearing a red hat and a blue scarf at the same time. This helps scientists understand how the virus is evolving as a complete package.

Why is this better than the old tools?

• Speed: It's 5 to 8 times faster than the current standard (VADR).
• Accuracy: Because it asks humans to double-check the maps, it makes fewer mistakes.
• Completeness: It handles viruses that are made of multiple pieces (like Influenza) or have complex instructions (like SARS-CoV-2) much better than anyone else.

The Bottom Line

VICAST is a bridge. It connects the messy, raw data from a lab to a clear, accurate story about how a virus is evolving. It combines the speed of a computer with the wisdom of a human expert, ensuring that scientists don't just see that the virus changed, but understand why and how it changed.

The creators even used AI to help write the user manuals and test the software, but the final decisions and the "detective work" were done by real scientists to ensure everything is 100% accurate.

Technical Summary

1. Problem Statement

Cultured virus passage studies are essential for understanding viral evolution, attenuation, and host adaptation. However, analyzing genomic data from these studies faces two distinct computational challenges that existing tools fail to address simultaneously:

• Inadequate Functional Annotation: Existing automated pipelines (e.g., VADR, VIGOR4) excel with well-characterized viruses but struggle with novel or poorly annotated genomes. Crucially, they often fail to resolve polyproteins into mature functional proteins, making it impossible to determine the specific biological impact of mutations (e.g., distinguishing a mutation in a specific protease domain versus a generic polyprotein region).
• Insensitivity to Low-Frequency Variants: Clinical variant calling pipelines prioritize consensus sequences, often filtering out low-frequency variants (3–50% allele frequency). These sub-consensus variants are biologically critical in passage studies as they represent the "leading edge" of viral evolution and adaptation but are invisible to standard consensus-based analysis.
• Lack of Integrated Workflows: Current tools separate annotation and variant calling, lacking features for contamination screening, haplotype reconstruction, and manual curation checkpoints necessary for high-quality research.

2. Methodology

VICAST (Viral Cultured-virus Annotation and SnpEff Toolkit) is an integrated software suite designed to bridge genome annotation curation and variant analysis. It is distributed via Docker and Conda and supports High-Performance Computing (HPC) environments. The toolkit consists of two primary components:

A. VICAST-annotate (Genome Annotation Curation)

This module implements a semi-automated workflow with four distinct pathways to handle diverse genome qualities:

1. Pre-built Database Check: Queries existing SnpEff databases.
2. GenBank Parsing: Processes genomes with complete CDS features and polyprotein mat_peptide annotations.
3. BLASTx Homology Search: Annotates novel/poorly characterized genomes using homology searches, generating TSV files for manual review.
4. Segmented Genome Handling: Concatenates multi-segmented viruses (e.g., Influenza) into a unified SnpEff database while preserving segment-specific annotations.

Key Feature: The workflow includes mandatory manual curation checkpoints. Researchers must review and verify gene names, coordinates, and features before database construction, ensuring domain expertise resolves ambiguities like ribosomal frameshifts or cleavage sites.

B. VICAST-analyze (Variant Calling and Annotation)

This module follows a nine-step, QC-first workflow:

1. Read Preparation: Quality control (fastp) and statistics.
2. Alignment & Calling: Mapping with BWA-MEM2 and low-frequency variant calling using lofreq (optimized for sub-consensus sensitivity).
3. Coverage Profiling: Diagnostic depth analysis.
4. Contamination Screening: A unique module performing de novo assembly (MEGAHIT) followed by BLAST screening against a curated database of 18,804 viral/microbial sequences. Contaminants are classified by query coverage (Confirmed ≥80%, Potential 50–80%).
5. Two-Stage Filtering: Applies empirically validated thresholds:
   • Dominant variants: ≥1% frequency, ≥200× depth, quality ≥1000.
   • Low-frequency variants: 3–50% frequency, ≥200× depth, quality ≥100.
6. Functional Annotation: Integration with SnpEff for effect prediction, specifically handling polyproteins and multi-segmented genomes.
7. Post-Processing: Generates consensus genomes and reconstructs low-frequency haplotypes using a BAM-level read co-occurrence module. This module analyzes read pairs to determine physical linkage between variants within 500 bp, validating haplotypes without long-read sequencing.

3. Key Contributions

• Integrated Workflow: Unifies annotation curation, contamination screening, and low-frequency variant calling into a single pipeline.
• Polyprotein Resolution: Provides protein-level annotations for polyproteins (e.g., SARS-CoV-2 ORF1ab, Dengue, Chikungunya), resolving mutations to specific mature peptides (e.g., nsp5, NS3 protease) rather than generic polyprotein regions.
• Curated Databases: Distributes 27 pre-built SnpEff databases, including novel, manually curated annotations for Chikungunya virus (NC_004162.2) that resolve mature peptide boundaries absent in NCBI references.
• Haplotype Reconstruction: Introduces a read co-occurrence module to validate physical linkage of low-frequency variants, enabling the reconstruction of quasispecies haplotypes from short-read Illumina data.
• Contamination Detection: Early-stage identification of adventitious viruses and microbes (e.g., Mycoplasma, fungi) via de novo assembly, preventing wasted analysis on compromised samples.

4. Results

• Validation: VICAST was validated on three distinct datasets:
  • SARS-CoV-2: Correctly resolved ORF1ab variants to specific nonstructural proteins (e.g., nsp5, nsp12) and identified key drug-resistance mutations.
  • Dengue Virus 2: Successfully detected low-frequency variants in structural (E protein) and nonstructural (NS3/NS5) domains.
  • Influenza A H1N1: Unified analysis of eight segments, correctly distinguishing spliced products (M2 vs. M1) and frameshifted products (PA-X).
• Benchmarking vs. VADR:
  • Speed: VICAST processed genomes 5.6–8.1 times faster than VADR (e.g., 3.5s vs. 28.4s for SARS-CoV-2).
  • Memory: VICAST required <100 MB peak memory, whereas VADR recommends up to 64 GB for large genomes.
  • Accuracy: While VADR had slightly higher automated accuracy (98.8% vs. 97.2%), VICAST's accuracy exceeded 99% after the mandatory manual curation step.
• Haplotype Validation: In SARS-CoV-2 and Dengue/Chikungunya co-infection samples, the co-occurrence module showed >99% linkage for variants on the same haplotype and <1% for unlinked variants, confirming the reliability of short-read phasing.

5. Significance

VICAST addresses a critical gap in viral genomics by recognizing that annotation quality dictates variant interpretation. By coupling semi-automated annotation with mandatory human curation, it ensures that low-frequency variants are interpreted in the correct functional context (e.g., specific protein domains).

• Research Impact: It enables rigorous analysis of viral evolution in passage studies, allowing researchers to track adaptive mutations, drug resistance, and attenuation mechanisms with high sensitivity.
• Community Resource: The distribution of curated databases (especially for Chikungunya) fills a void in public repositories, providing immediate functional annotation capabilities for the virology community.
• Practical Utility: The integrated contamination screening and haplotype reconstruction features make it a robust tool for ensuring data integrity and understanding viral population dynamics without the need for expensive long-read sequencing.

VICAST is freely available at https://github.com/mihinduk/VICAST.