Portello: Making global assembly more effective for rare-disease whole genome sequencing

Portello is a novel assembly-based mapping tool that transfers read alignments from de novo assembly contigs to a standard reference genome, thereby unifying assembly and reference-based inferences to improve variant calling accuracy, enable direct read-level review, and facilitate the interpretation of complex genomic regions for rare-disease whole genome sequencing.

The Gist

Imagine you are trying to solve a massive, intricate jigsaw puzzle of a human genome to find the tiny, hidden piece causing a rare disease.

For a long time, scientists have had two main ways to do this:

1. The "Map" Method: They take the puzzle pieces (DNA reads) and try to force them onto a pre-existing, generic map of a human genome. It's like trying to fit a unique, custom-shaped puzzle piece onto a standard map. If the piece is weird or the map has a typo, the piece gets stuck in the wrong spot, or you miss it entirely.
2. The "Build" Method: They ignore the map and build the puzzle from scratch (de novo assembly). This is great because it captures the unique shape of the patient's DNA perfectly. But, once the puzzle is built, it's hard to look at the original pieces again to double-check your work, and it's hard to compare your custom puzzle to the standard map to see what's different.

Enter "Portello."

Think of Portello as a super-smart translator and tour guide that bridges the gap between these two methods. Here is how it works, using simple analogies:

1. The Problem: The "Lost in Translation" Issue

When scientists build a custom genome puzzle (the assembly), they often lose the ability to easily see the "footprints" of the original DNA pieces. It's like building a house and then throwing away the blueprints and the list of materials. If you find a weird crack in the wall, you can't easily check the original delivery receipts to see if the brick was defective or if you just laid it wrong.

Also, if you build a custom house, it's hard to tell the insurance adjuster (the standard medical tools) exactly where the rooms are located on their standard city map.

2. The Portello Solution: The "Two-Step Relay"

Portello solves this by doing a clever two-step relay race:

• Step 1: The Local Guide. First, it takes the DNA pieces and maps them to the custom puzzle you just built. Since the pieces fit perfectly here, the map is accurate.
• Step 2: The Global Translator. Then, it takes that perfect custom puzzle and maps it onto the standard city map (the reference genome).

Because Portello knows exactly how the pieces fit the custom puzzle, and exactly how the custom puzzle fits the city map, it can translate the location of every single DNA piece directly onto the city map.

The Magic Result: You now have a map that looks exactly like the standard city map (so all your usual tools work), but the "traffic" (the DNA pieces) is flowing in the correct lanes because it was guided by the custom puzzle.

3. Why This Matters (The Real-World Impact)

• Fewer Mistakes (The "False Alarms"):
  In the old "Map" method, scientists often missed tiny errors (False Negatives) or thought they saw errors where there were none (False Positives).

  • The Analogy: Imagine trying to read a blurry sign from a moving car. Portello is like stopping the car, cleaning the windshield, and reading the sign from a stationary, high-definition camera.
  • The Result: The paper shows Portello cut the number of missed errors by nearly 50%. It found the "bad apples" that the old method threw away.

• Seeing the Invisible (Complex Regions):
  Some parts of the genome are like a "hall of mirrors" (repetitive regions). In the old method, DNA pieces would bounce around and get lost.

  • The Analogy: If you try to map a maze by looking at a flat photo, you get confused. Portello builds a 3D model of the maze first, then tells you exactly where you are in the flat photo.
  • The Result: It made it possible to clearly see "Copy Number Variations" (extra or missing chunks of DNA) in these messy areas, which is crucial for diagnosing diseases.

• The "Family Album" (Phasing):
  Humans have two copies of every chromosome (one from mom, one from dad). Sometimes, the old method mixes them up, like putting a photo of your mom next to your dad's name.

  • The Analogy: Portello acts like a family photo organizer. It looks at the custom puzzle, sees which pieces belong to "Mom's side" and which belong to "Dad's side," and then tags the DNA pieces on the standard map accordingly.
  • The Result: Doctors can now see exactly which version of a gene is broken, which is vital for understanding genetic diseases.

The Bottom Line

Portello doesn't just build a better puzzle; it makes the puzzle playable with all the standard tools doctors already use.

It takes the best of the "Build from Scratch" method (high accuracy, captures unique DNA) and combines it with the best of the "Standard Map" method (easy to use, easy to visualize). This means doctors can find the causes of rare diseases faster, with fewer mistakes, and with a clearer picture of what's actually happening in a patient's DNA.

In short: Portello is the bridge that lets us use our most powerful new tools without leaving behind the old tools we rely on.

Technical Summary

1. Problem Statement

While de novo assembly of long-read HiFi data offers superior sensitivity for detecting rare disease variants (especially those divergent from the reference genome), it remains underutilized in clinical workflows due to several critical bottlenecks:

• Lack of Read-Level Context: It is difficult to review the raw read evidence (basecall qualities, methylation, mosaic variants) supporting specific assembly-based variant calls.
• Inconsistent Variant Representation: Conflicts often arise between variant calls derived from assembly contigs and those from conventional reference-based mapping pipelines. Resolving these discrepancies (e.g., a region called as a deletion by assembly but containing small variants by mapping) is challenging.
• Visualization and Interpretation: Standard genome browsers and visualization tools are optimized for reference-based alignments, making it hard to visualize assembly-based findings in a familiar context.
• Phasing Complexity: Integrating phasing information from partially-phased assemblies with standard read-based phasing is non-trivial.

2. Methodology: The Portello Approach

Portello is a computational tool designed to transfer read mappings from a sample's de novo assembly contigs onto a standard reference genome (e.g., GRCh38). It acts as a bridge, creating "assembly-based read mappings" that retain the accuracy of the assembly while conforming to standard reference coordinates.

Core Workflow:

1. Input Generation:
   • Read-to-Contig: Reads are mapped back to their own de novo assembly contigs (typically using pbmm2).
   • Contig-to-Reference: The assembly contigs are mapped to the standard reference genome (typically using minimap2).
2. Mapping Transfer (Composition):
   • Portello composes the two alignment steps (Read $\to$ Contig $\to$ Reference) to generate a new alignment for each read directly against the reference.
   • It correctly handles split alignments and circular contigs (e.g., mitochondria).
3. Pre-processing & Optimization:
   • Repeated Matches Trimming: Ensures each base in a contig maps to the reference only once, preventing artificial coverage spikes in repetitive regions.
   • Collinear Segment Joining: Merges co-linear alignment segments that were artificially split by the reference mapper (e.g., due to Z-drop criteria), improving small variant calling and visualization.
   • Left-Shifting Indels: Normalizes indel positions to ensure consistency, particularly when contigs map in reverse orientations.
4. Phasing Annotation:
   • Portello annotates reads with haplotype tags (HP) and phase set tags (PS) based on the contig they originated from.
   • It supports both fully-phased inputs (e.g., from parental scaffolding) and partially-phased inputs (dual assemblies with potential switch errors) by evaluating read support across heterozygous variant pairs.

3. Key Contributions

• Unified Workflow: Enables the integration of assembly-based structural variant (SV) calls with standard reference-based small variant calling tools (like DeepVariant) on a single, consistent genomic view.
• Enhanced Visualization: Allows researchers to visualize assembly-based variants alongside raw read evidence (methylation, base quality) in standard genome browsers.
• Improved CNV Detection: Facilitates the detection of copy number variations (CNVs) in complex regions (e.g., segmental duplications) by leveraging the "decoy" effect of the assembly to resolve mapping ambiguities.
• Open Source Tool: The software is released as a pre-compiled binary and source code on GitHub, designed to work with existing PacBio pipelines.

4. Results

The authors evaluated Portello using HiFi data from HG002, NA12878, and NA21886.

• Small Variant Calling Accuracy:
  • Using DeepVariant on Portello-remapped reads significantly reduced errors compared to conventional pbmm2 mapping.
  • HG002: Reduced false negatives by 52% and total basecall errors by 47%.
  • NA12878: Reduced false negatives by 55% and total basecall errors by 29%.
  • Crucially, these gains in recall were achieved with almost no increase in false positives.
• Phasing Accuracy:
  • For partially-phased assemblies, Portello achieved a block NG50 of 334,357 bases with a low switch error count (274 switches, 67 flips) when benchmarked against the GIAB T2T truth set.
• Complex Loci & CNV Interpretation:
  • In a case study of a 225 kb duplication in a segmental duplication region (NA21886), conventional mapping showed extreme coverage irregularities (spikes up to 12,000x) and dropouts, making interpretation impossible.
  • Portello mapping revealed a clean 3-fold coverage region corresponding to the duplication, allowing for high-confidence CNV calling and clear visualization of the underlying assembly structure.

5. Significance

Portello represents a paradigm shift in rare-disease genomics by making de novo assembly a practical, routine component of clinical analysis rather than a niche, hard-to-interpret method.

• Clinical Utility: It solves the "interpretability gap," allowing clinicians to trust assembly-based findings by visualizing the supporting read evidence directly on the reference genome.
• Hybrid Analysis: It enables "best-of-both-worlds" pipelines where high-confidence SVs from assembly are combined with high-precision small variants from standard callers, all unified under one coordinate system.
• Future-Proofing: As long-read sequencing becomes standard for rare diseases, Portello provides the necessary infrastructure to fully leverage the sensitivity of de novo assembly without sacrificing the compatibility of existing bioinformatics ecosystems.