NOHIC: A PIPELINE FOR PLANT CONTIG SCAFFOLDING USING PERSONALIZED REFERENCES FROM PANGENOME GRAPHS

The paper introduces noHiC, a reference-guided scaffolding pipeline that utilizes a pangenome graph-based algorithm to generate personalized synthetic references, enabling the rapid and accurate assembly of plant genomes into highly contiguous structures without the need for costly Hi-C sequencing.

The Gist

The Big Picture: Solving the "Jigsaw Puzzle" Problem

Imagine you have just finished a massive, complex jigsaw puzzle, but the pieces are all mixed up in a pile. You know what the final picture should look like because you have a picture on the box (the Reference Genome).

In the world of plant genetics, scientists often have the "pieces" (DNA sequences called contigs) but need to put them in the right order to see the whole plant's genetic blueprint. Usually, they use a "box picture" from a famous, well-known plant to guide them.

The Problem:
If the plant you are studying is a bit different from the famous one on the box (maybe it's a wild cousin or a different variety), trying to force your pieces to fit the old box picture causes problems. You might break a piece that actually fits, or force two pieces together that don't belong. This is called Reference Bias. It's like trying to force a square peg into a round hole just because the instruction manual says "square peg goes here."

Also, the traditional way to fix these puzzles involves taking expensive, high-tech photos of the puzzle while it's being built (called Hi-C sequencing). This is slow, costly, and requires special equipment.

The Solution:
The authors created a new tool called noHiC. It's a "smart assistant" that helps you assemble your plant puzzle without needing those expensive photos. Instead of using just one old box picture, it builds a custom, personalized box picture specifically for your plant.

---

How noHiC Works: The Three-Step Magic

The pipeline works like a three-stage assembly line:

1. The Cleanup Crew (nohic-clean)

Before you start building, you have to clean your workspace.

• The Analogy: Imagine your puzzle pieces are sitting on a table covered in dust, cat hair, and maybe a few pieces from a different puzzle (like a cat toy).
• What it does: This step scans your DNA pieces and throws away anything that isn't the plant (like bacteria or viruses) and removes the "plastic wrap" (sequencing adapters) stuck to the pieces. It ensures you are only working with the plant's actual DNA.

2. The Custom Blueprint Generator (nohic-refpick)

This is the most magical part of the paper.

• The Analogy: Imagine you have a giant library containing the blueprints of 48 different versions of the same house (a Pangenome Graph). Some have blue roofs, some have red doors, some have extra garages.
• The Problem: If you try to build your specific house using just one of those blueprints, you might get it wrong if your house is a mix of features from all of them.
• The Solution: The nohic-refpick script acts like a super-smart architect. It looks at your specific house (your target plant) and says, "Okay, for the roof, I'll take the design from House #3. For the door, I'll take House #12. For the garage, House #45."
• The Result: It stitches together the best 10,000-base-pair chunks from all those different blueprints to create a Synthetic Reference (Synref). This new blueprint is a perfect genetic match for your specific plant, even though it never existed before. It combines the best of everyone in the family tree.

3. The Assembly Line (nohic-asm or ntJoin)

Now that you have clean pieces and a perfect custom blueprint, you start building.

• The Analogy: You lay out your pieces and snap them together according to your custom blueprint.
• The "Smart" Fix: Sometimes, the pieces are still a little crooked. The tool checks for "glue errors" (misassemblies) and fixes them.
• The Fast Track: The paper also shows you can swap the slow, careful assembly line for a "turbo-charged" one (called ntJoin) if you are in a rush. Even with the fast method, using your Custom Blueprint (Synref) still gives you a much better result than using the old, generic box picture.

---

Why is this a Big Deal?

1. It Saves Money: You don't need the expensive, time-consuming Hi-C sequencing photos anymore. You can just use the DNA data you already have.
2. It's Fairer: By creating a "personalized" reference, it stops the bias where scientists accidentally erase unique genetic traits just because they don't match the "standard" reference. It preserves the unique quirks of your specific plant.
3. It's Reusable: Once you build that giant library of blueprints (the Pangenome Graph) for a species (like Sorghum or Barley), you can use it forever to build custom blueprints for any new plant of that species you discover. You don't have to start from scratch every time.
4. It Works Everywhere: The team tested this on four very different plants (a tiny weed, a grass, a bean, and a barley). In almost every case, the custom blueprint produced a more complete, less broken puzzle than the standard reference.

The Bottom Line

noHiC is like upgrading from using a generic, one-size-fits-all instruction manual to having a tailor-made guide that knows exactly how your specific plant is built. It lets scientists assemble high-quality plant genomes faster, cheaper, and more accurately, ensuring that the unique genetic story of every plant is told correctly.

Technical Summary

1. Problem Statement

While high-quality plant genome assemblies are increasingly achievable using long-read sequencing (PacBio HiFi, ONT), the process of scaffolding (ordering and orienting contigs into chromosomes) remains challenging.

• Hi-C Limitations: Reference-free scaffolding using Hi-C data is costly, computationally intensive, and requires complex library preparation.
• Reference-Guided Limitations: Reference-guided scaffolding is cheaper and faster but suffers from reference bias. When a target genome is divergent from the single reference used, alignment errors occur, leading to:
  • Failure to reconstruct divergent regions.
  • Misassemblies (incorrect joins or breaks).
  • Reduced representation of genetic diversity.
• Existing Multi-Reference Tools: Tools like Ragout2, ntJoin, and Multi-CSAR attempt to use multiple references but are inefficient for large numbers of references (>10), require complex alignment updates (e.g., HAL/MAF files), or need manual weight optimization for each new target.

2. Methodology: The noHiC Pipeline

The authors introduce noHiC, a reference-guided scaffolding pipeline designed to mitigate reference bias by generating a personalized synthetic reference (synref) from a pangenome graph. The pipeline consists of four modular sub-scripts:

A. nohic-clean (Contaminant Removal)

• Function: Removes non-target sequences (bacteria, fungi) and organellar DNA (mitochondria/chloroplasts) from raw contigs.
• Tools: Uses Kraken2 and Taxonkit for taxonomic classification and BLASTn for organelle removal.

B. nohic-refpick (Personalized Reference Generation)

• Core Innovation: This is the distinct feature of noHiC. It generates a synref tailored to the target genome.
• Algorithm: It utilizes the haplotype sampling algorithm (Sirén et al.) on a pangenome graph (.gbz format).
  1. The graph is partitioned into 10-kb blocks.
  2. Graph-unique k-mers (k-mers occurring once in the graph) are extracted.
  3. These k-mers are queried against the target's sequencing reads (HiFi) to determine presence/absence and zygosity.
  4. The algorithm selects the best-fitting haplotype blocks to construct a single, artificial reference genome that is genetically closest to the target.
• Gap Patching: An optional step (--patch) fills gaps in the synref using a high-quality donor genome (e.g., a T2T assembly) via minimap2 and GPatch.

C. nohic-asm (Correction and Scaffolding)

• Function: Corrects misassemblies and scaffolds contigs.
• Workflow:
  1. Chimeric Breaking: Uses CRAQ to break contigs based on clipped reads.
  2. Small Error Correction: Uses Inspector to fix base substitutions, inversions, etc.
  3. Reference-Guided Correction: Uses RagTag to break remaining chimeric contigs based on alignment to the synref (or conventional reference).
  4. Scaffolding: Orients and joins contigs using RagTag.
  5. Gap Closing: Uses TGSGapcloser.
• Presets: Offers five correction presets ranging from "relaxed" (draft, luck) to "strict" (standard, aggressive, raw), allowing users to balance contiguity against structural accuracy.

D. nohic-eval (Quality Assessment)

• Metrics: Calculates contiguity (N50, auN, gap count), gene completeness (BUSCO), and structural correctness (R-AQI, S-AQI, QV).
• Visualization: Generates error maps and dot plots to visualize synteny and misassemblies.

3. Key Contributions

1. Synref Generation: The first application of pangenome graph-based haplotype sampling to generate a single, best-fit reference for contig scaffolding. This integrates genetic information from up to 48 references without requiring multiple whole-genome alignments.
2. Bias Mitigation: By using a synref that is genetically closer to the target than any single conventional reference, the pipeline significantly reduces reference bias during the correction phase.
3. Modularity & Flexibility: The pipeline can be used as a full suite or combined with ultra-fast scaffolders like ntJoin. The nohic-refpick script can generate synrefs for use with any alignment-based or alignment-free scaffolder.
4. Reusability: Pangenome graphs can be constructed once and reused to generate personalized references for multiple new target genomes within a species, eliminating the need to re-align to multiple references for every new project.

4. Results

The pipeline was tested on four plant species (Arabidopsis thaliana, Sorghum virgatum, Glycine max, Hordeum vulgare) and three in-house sorghum accessions.

• Genetic Closeness: Phylogenetic analysis (NJ trees) confirmed that synrefs generated by nohic-refpick were consistently the most genetically similar to the target genomes compared to conventional references or other assemblies in the pangenome.
• Contiguity Preservation:
  • Under the strict "standard" correction preset, synrefs significantly reduced false contig breaking compared to conventional references.
  • In sorghum, synref-guided assemblies showed up to 65% higher contig auN (a measure of contiguity) and 31% fewer contigs than those guided by the closest conventional reference (BTx623 v5).
  • In barley (H. vulgare), synrefs reduced contig and gap numbers by >50% compared to the Morex v3 reference.
• Structural Accuracy:
  • Assemblies guided by synrefs showed strong synteny with public Hi-C-based or manually curated controls.
  • Conventional references often introduced false interchromosomal translocations (e.g., in Arabidopsis and Sorghum), which were corrected or minimized when using synrefs.
• Performance with ntJoin: When nohic-refpick was combined with the ultra-fast scaffolder ntJoin:
  • Runtime was reduced drastically (e.g., from ~24 hours to ~11 minutes for barley).
  • Despite ntJoin lacking read-based validation, synref-guided assemblies still outperformed conventional reference-guided assemblies in contiguity and structural correctness (QV, R-AQI).
• Limitations: In cases where the target genome is very close to a high-quality, gapless conventional reference (e.g., Glycine max), the advantage of synrefs was minimal under relaxed correction settings.

5. Significance

• Cost-Effective Genomics: noHiC enables the assembly of high-quality, chromosome-scale plant genomes without the high cost and complexity of Hi-C sequencing.
• Scalability: It solves the bottleneck of multi-reference scaffolding by converting complex multi-reference alignment problems into a single alignment against a personalized synref.
• Broad Applicability: The pipeline is effective across diverse plant species with varying genome sizes (135 Mb to 4.2 Gb) and ploidy levels.
• Future-Proofing: As pangenome graphs become standard in plant genomics, noHiC provides a direct pathway to utilize this rich genetic diversity for improving individual genome assemblies, ensuring that "reference bias" does not obscure true genetic variation in large-scale population studies.

The pipeline is publicly available at https://github.com/andyngh/noHiC.