Accounting for Defective Viral Genomes in viral consensus genome reconstruction, application to influenza virus

The paper introduces DIPScan, a scalable Nextflow workflow that accurately detects deletion-containing viral genomes (DelVGs) in influenza sequencing data and corrects their interference to ensure the generation of precise consensus genomes for epidemic surveillance.

The Gist

The Big Picture: The "Broken Copy" Problem

Imagine you are a librarian trying to reconstruct the perfect, original version of a famous novel (the Virus Genome) based on thousands of photocopies (the Sequencing Reads) that people have brought in.

Usually, this is easy. You look at all the photocopies, see what the majority say, and write down the "Consensus" (the most common version).

But here's the catch: Some of these photocopies are broken. They have huge chunks of pages missing (deletions) and sometimes have weird, random scribbles added in the margins (mutations). These are called Defective Viral Genomes (DelVGs).

The problem? In a viral infection, these "broken copies" can sometimes outnumber the "perfect copies." If your librarian software just looks at the majority vote, it might accidentally write the broken, missing pages into the final book, thinking that's what the original story was supposed to be. This creates a chimeric (mixed-up) version of the virus that doesn't actually exist in nature, leading to bad science and bad public health decisions.

The Solution: Meet DIPScan

The authors of this paper created a new tool called DIPScan (Defective Interfering Particle Scanner). Think of DIPScan as a super-smart editor that doesn't just count words; it understands the structure of the book.

Here is how DIPScan works, step-by-step:

1. The "Gap Detector" (Finding the Broken Copies)

Standard tools look for single-letter typos. DIPScan looks for gaps.

• Analogy: Imagine reading a sentence: "The quick brown fox jumps over the lazy dog."
• A broken copy might look like: "The quick brown fox [GAP] lazy dog."
• DIPScan notices that the middle is missing. It doesn't just ignore it; it flags it. It looks for "split reads"—pieces of the photocopy that don't connect, indicating a massive deletion.

2. The "Vote Counter" (Estimating the Mix)

Once DIPScan finds the broken copies, it asks: "How many broken copies are there compared to the perfect ones?"

• Analogy: If 90% of the photocopies are missing the middle chapter, but 10% are perfect, the "perfect" version is the minority.
• DIPScan uses complex math to figure out the exact ratio. If the broken copies are the majority, it knows the standard "majority vote" method will fail.

3. The "Editor" (Fixing the Consensus)

This is the magic part. DIPScan looks at the final draft of the virus genome and asks: "Is this weird mutation coming from the perfect virus, or is it just a scribble on the broken copies?"

• Scenario A: If the mutation is on the broken copies, DIPScan says, "Delete that!" and replaces it with a question mark (an 'N') or the correct letter from the perfect virus.
• Scenario B: If the mutation is on the perfect virus, it keeps it.
• Result: The final book (Consensus Genome) is clean, accurate, and represents the real virus, not the broken noise.

Why This Matters (The "Why Should I Care?")

The paper tested this tool on Influenza (the Flu) because flu viruses are notorious for making these broken copies.

• The Scale: They looked at hundreds of real patient samples. They found that about 30% of flu samples had these broken copies.
• The Risk: Without DIPScan, scientists might have been publishing "fake" flu virus sequences that looked like they had mutations they didn't actually have. This could mess up vaccine development or drug resistance tracking.
• The Success: DIPScan caught these errors with 99% accuracy. It found the broken copies that human experts missed and fixed the genome sequences automatically.

The "Hotspots" Discovery

While fixing the genomes, DIPScan also noticed a pattern. The breaks in the viral "books" didn't happen randomly. They happened in specific "danger zones" near the beginning and end of the genetic segments.

• Analogy: It's like realizing that every time someone tears a page out of a specific type of notebook, they always tear it out of the first 10 pages or the last 10 pages.
• This helps scientists understand how the virus breaks itself, which could lead to new ways to stop it.

Summary

DIPScan is a digital "spell-checker" for virus genomes that is smart enough to know the difference between a real mutation and a broken, missing page. By cleaning up the data, it ensures that when we track viruses like the Flu, we are looking at the real enemy, not a distorted reflection.

The tool is now being used routinely at the Pasteur Institute in Paris to keep our global virus surveillance accurate.

Technical Summary

1. Problem Statement

In viral epidemic surveillance, generating accurate consensus genomes from Next-Generation Sequencing (NGS) data is critical for tracking mutations and predicting viral evolution. However, standard consensus reconstruction workflows often fail to account for Deletion-containing Viral Genomes (DelVGs).

• The Issue: DelVGs are defective viral particles containing truncated or rearranged versions of the genome. In clinical samples, DelVGs can sometimes outnumber full-length genomes.
• The Consequence: Because DelVGs often cover the terminal regions of the genome (where deletions occur internally), they create high coverage at the ends and low coverage in the center. If DelVGs carry unique mutations absent in the full-length genome, standard consensus algorithms (which typically select the majority nucleotide) erroneously incorporate these DelVG-specific mutations into the final consensus.
• The Risk: This leads to "chimeric" consensus sequences that misrepresent the true circulating viral population, potentially introducing frameshifts or premature stop codons, and compromising downstream surveillance and phylogenetic analysis.
• Current Limitations: Existing tools for detecting defective genomes (e.g., ViReMa, DG-Seq, VODKA2) are often designed for research rather than routine surveillance, lack scalability, or fail to correct the consensus sequence itself.

2. Methodology: DIPScan

The authors developed DIPScan, a Nextflow-based bioinformatics workflow designed to automatically detect DelVGs and correct consensus sequences. The workflow consists of seven primary computational steps:

1. Read Mapping (Split-Read Strategy):

   • Step 1: Reads are mapped to a reference genome using BWA-MEM2. Reads with <8 clipped nucleotides are retained as standard alignments.
   • Step 2: Unmapped reads and those with significant clipping (≥8 nucleotides) are extracted and realigned using STAR (originally designed for splicing) to identify "split reads" spanning large deletions.
   • The outputs are merged into a single BAM file.

2. Extraction of Deletion Boundaries:

   • Reads containing "skipped regions" (denoted by 'N' in the CIGAR string) >150 nucleotides are identified.
   • Start and end coordinates of these deletions are consolidated into a non-redundant set.

3. Metric Computation:

   • For each deletion boundary, the tool calculates:
     • Supporting reads: Count of reads spanning the exact deletion.
     • Split/Total frequency: Relative abundance of the deletion.
     • Deletion start/end ratios: Local coverage changes at breakpoints to distinguish true deletions from noise.

4. Breakpoint Selection:

   • Filters are applied to remove noise (e.g., requiring >100 supporting reads, total frequency > expected minimum, and specific coverage ratios <1).

5. DelVG Proportion Estimation:

   • The workflow models the sample as a mixture of full-length genomes and $d$ distinct DelVGs.
   • It solves a system of linear equations using Non-Negative Least Squares (NNLS) to estimate the relative abundance of each DelVG and the full-length genome based on observed median coverages across genomic regions and split-read counts.
   • A sample is flagged as "DelVG-positive" if the sum of DelVG proportions exceeds 50%.

6. Consensus Correction:

   • Region Definition: The tool identifies "correction regions" (start and end of the genome) where DelVG mutations are likely to dominate.
   • Mutation Selection: It identifies variants within these regions using iVar.
   • Nucleotide Resolution: For each mutated position, DIPScan determines the correct nucleotide based on:
     • The estimated proportion of the full-length genome ($p_{fl}$).
     • The frequency of the mutation ($f_{mi}$).
     • Logic:
       • If $p_{fl}$ is low (<2%), the position is masked with 'N'.
       • If the full-length genome is dominant, the consensus retains the full-length nucleotide.
       • If the DelVG is dominant, the tool attempts to mathematically deconvolve the mixture (solving a "Multiple Subset Sum" problem) to assign the correct full-length nucleotide. If ambiguous, it masks the position with 'N'.

3. Key Contributions

• Novel Workflow: DIPScan is the first tool specifically designed to not only detect DelVGs but also automatically correct the consensus sequence in routine surveillance pipelines.
• Scalability: Implemented in Nextflow, making it reproducible, containerized, and scalable for high-throughput processing (e.g., thousands of samples).
• Quantitative Estimation: It provides a robust mathematical estimation of DelVG vs. full-length genome proportions, rather than just binary detection.
• Integration: It is already integrated into the routine pipeline of the National Reference Center (NRC) for Respiratory Viruses at the Institut Pasteur.

4. Results

The tool was validated using two datasets: 110 simulated datasets and 551 real patient-derived influenza samples.

• Breakpoint Detection Accuracy:
  • Compared to ViReMa, DG-Seq, and VODKA2, DIPScan achieved 100% precision and 96.9% F1-score.
  • Other tools produced high false-positive rates (e.g., DG-Seq had 83% FDR), whereas DIPScan missed only 6 true breakpoints (mostly low-frequency).
• Proportion Estimation:
  • In simulated data, DIPScan showed a Pearson correlation of 0.99 between estimated and true DelVG proportions.
  • In real data, estimates correlated highly ($r=0.96$) with empirical mutation frequencies used as a proxy for ground truth.
• Consensus Correction:
  • In simulations, DIPScan correctly handled 97.2% of DelVG-specific mutations (either correcting them to the full-length base or masking them).
  • In real data validation (25 unambiguous positions), 96% were correctly handled (changed, retained, or masked).
• Real-World Application:
  • Applied to 551 influenza samples (A/H1N1pdm, A/H3N2, B/Victoria), DIPScan identified DelVGs in ~30% of samples, primarily affecting polymerase segments (PB1, PB2, PA).
  • It revealed subtype-specific breakpoint hotspots near the 5' and 3' ends of the polymerase segments, confirming previous biological hypotheses on a large scale.

5. Significance

• Data Integrity: DIPScan prevents the contamination of public databases (like GISAID) with erroneous consensus sequences caused by defective genomes, ensuring that surveillance data accurately reflects circulating viral strains.
• Public Health Impact: By improving the accuracy of consensus genomes, the tool enhances the detection of true mutations of concern (e.g., drug resistance, immune escape) and improves phylogenetic tracking.
• Routine Viability: The tool bridges the gap between research-grade detection of defective genomes and the need for automated, high-throughput processing in national surveillance centers.
• Future Outlook: The authors plan to extend DIPScan to other pathogens (RSV, SARS-CoV-2) and refine it to detect other types of defective genomes (e.g., copy-back or rearrangements).