ITSxRust: ITS region extraction with partial-chain recovery and structured diagnostics for long-read amplicon sequencing

ITSxRust is a high-performance Rust-based tool that significantly improves the throughput, accuracy, and robustness of ITS region extraction for long-read fungal amplicon sequencing by integrating efficient HMMER searches with a novel partial-chain recovery strategy and structured diagnostics.

The Gist

The "Fungal ID Card" Problem

Imagine you are a librarian trying to organize a massive library of fungal DNA. In this library, every book (a piece of DNA) has a specific chapter called the ITS region. This chapter is the "fingerprint" or "ID card" that tells you exactly what kind of fungus you are looking at.

However, the books are messy. The ITS chapter is sandwiched between two very boring, identical chapters (the "flanks") that look the same in almost every fungus. To read the ID correctly, you have to cut out only the middle chapter.

For a long time, scientists used two main tools to do this cutting: ITSx and ITSxpress.

• ITSx is like a careful, old-school librarian. It finds the chapters perfectly but moves very slowly.
• ITSxpress is like a speed-reading robot. It's incredibly fast, but it tries to group identical books together to save time. The problem? When the books are from long-read sequencing (a newer, faster technology that produces huge, messy books), the "robot" gets confused. Because every long-read book has tiny, unique scribbles (errors), the robot thinks they are all different and stops grouping them, causing it to throw away most of the books.

Enter ITSxRust: The High-Speed, Smart Scissors

The authors of this paper built a new tool called ITSxRust. Think of it as a high-speed, robotic librarian built with modern materials (Rust) that is specifically designed to handle these huge, messy long-read books.

Here is how it works, using simple analogies:

1. The "Four-Anchor" Strategy

Imagine the ITS region is a bridge. To know exactly where the bridge starts and ends, you need to find four specific landmarks (anchors) on both sides:

1. The end of the left bank.
2. The start of the bridge.
3. The end of the bridge.
4. The start of the right bank.

ITSxRust looks for all four landmarks. If it finds them, it cuts the bridge perfectly. This is the "Gold Standard."

2. The "Partial-Chain" Safety Net

Here is the genius part. In the messy world of long-read sequencing, sometimes a book is torn, and you can only find two of the four landmarks (e.g., you found the start of the bridge and the end of the bridge, but the left bank is missing).

Old tools would say, "I can't find all four, so I'm throwing this book away."
ITSxRust says, "Wait! Even with just two landmarks, I can still cut out the bridge safely."

This is called the Partial-Chain Fallback. It's like a safety net that catches the books that would otherwise fall into the trash can. In their tests, this feature saved 10,725 extra fungal IDs that other tools would have lost.

3. The "Smart Report"

If a book is too damaged to cut, ITSxRust doesn't just say "Error." It gives you a structured diagnostic report.

• Old tools: "Failed."
• ITSxRust: "Failed because the left bank landmark was missing. This suggests your DNA primer (the tool used to grab the DNA) might be cutting too deep into the wrong area."

This is like a mechanic telling you exactly why your car won't start, rather than just saying "Car broken."

The Results: Speed and Accuracy

The team tested this new tool on a massive dataset of 54,659 fungal reads (a huge library).

• Speed: ITSxRust was 4.6 times faster than the careful old librarian (ITSx). It finished the job in about 15 minutes, while ITSx took over an hour.
• Success Rate: It successfully extracted the ID cards from 75.3% of the books, beating ITSx (69.9%) and crushing ITSxpress (which only managed 41.4% because it got confused by the messy long reads).
• Accuracy: Even though it was fast, it didn't cut the ID cards wrong. The "cuts" were just as precise as the slow tools, meaning the fungi were identified correctly.

Why This Matters

In the world of science, time is money, and data is gold.

• If you use the old slow tool, you wait forever.
• If you use the fast-but-bad tool, you lose half your data.
• ITSxRust gives you the best of both worlds: it's fast, it saves data that others would throw away, and it tells you exactly what went wrong if something fails.

It's essentially upgrading the library from a slow, manual process to a modern, automated system that knows how to handle the messy reality of modern technology, ensuring no fungal ID is left behind.

Technical Summary

1. Problem Statement

The fungal Internal Transcribed Spacer (ITS) region (ITS1–5.8S–ITS2) is the standard DNA barcode for fungal taxonomy. As long-read sequencing technologies (e.g., Oxford Nanopore Technologies [ONT] and PacBio HiFi) become routine for metabarcoding, existing preprocessing tools face significant bottlenecks:

• Throughput and Robustness: Standard tools like ITSx and ITSxpress struggle with the volume and error profiles of long-read data.
• Partial Reads: Long-read amplicons often contain truncated sequences or missing conserved flanks (SSU/18S or LSU/28S), causing standard four-anchor detection methods to fail and discard valid data.
• Lack of Diagnostics: Existing tools provide limited feedback on why a read failed extraction, hindering troubleshooting of experimental parameters (e.g., primer design).
• Performance: ITSx is slow due to process spawning and text parsing overhead; ITSxpress relies on clustering which is ineffective for ONT data where unique error patterns prevent sequence deduplication.

2. Methodology

ITSxRust is a new extractor implemented in Rust, designed specifically for long-read workflows. Its architecture focuses on performance, robustness, and diagnostic transparency.

• Core Algorithm (Anchor Chain Model):
  • The tool uses HMMER (nhmmer) to search for profile-HMM hits against four conserved ribosomal anchors: SSU 3′end, 5.8S 5′start, 5.8S 3′end, and LSU 5′start.
  • Four-Anchor Chain: It attempts to construct a valid chain of all four anchors on a single strand. The chain with the highest sum-of-scores is selected.
  • Partial-Chain Fallback: If a full four-anchor chain cannot be formed (common in truncated reads), the tool attempts to extract subregions using two-anchor pairs (e.g., SSU 3′end + 5.8S 5′start for ITS1). This recovers reads missing one or two flanks while maintaining boundary quality defined by HMM hits rather than arbitrary read edges.
• Performance Optimizations:
  • Streaming I/O: Parses HMMER tblout output line-by-line to minimize memory usage.
  • Dereplication: Supports exact sequence deduplication to avoid redundant HMM searches (particularly effective for high-accuracy PacBio HiFi data).
  • Platform Presets: Includes specific parameter sets for ONT (relaxed length/E-value thresholds) and HiFi (stricter thresholds).
• Output and Diagnostics:
  • Generates structured TSV/JSONL reports per read containing anchor coordinates, scores, E-values, and confidence labels (Confident, Partial, or Ambiguous).
  • Provides a QC Summary (JSON) aggregating extraction counts and specific failure codes (e.g., "Missing SSU anchor"), enabling users to diagnose experimental issues.

3. Key Contributions

1. Partial-Chain Recovery: A novel fallback strategy that extracts subregions using two-anchor pairs when full chains are unavailable, significantly increasing extraction yield on truncated long-read datasets.
2. Structured Diagnostics: Unlike previous tools, ITSxRust provides granular failure reasons for every skipped read, allowing users to distinguish between biological issues (e.g., primer placement) and technical limitations.
3. Rust-Based Performance: Re-engineering the workflow in Rust reduces overhead from process spawning and text parsing, achieving significantly higher throughput than ITSx.
4. Platform Adaptability: Specific presets for ONT and HiFi data address the distinct error profiles and read lengths of these technologies.

4. Results

The tool was benchmarked on a real Oxford Nanopore dataset (54,659 reads) and simulated datasets (ONT and HiFi).

• Throughput:
  • ITSxRust was 4.6× faster than ITSx (15 minutes vs. 72 minutes).
  • While ITSxpress v2 was faster (7–8 minutes), it suffered from a massive drop in extraction yield on ONT data.
• Extraction Success:
  • Full ITS Extraction: ITSxRust (ONT preset) extracted 75.3% of reads, outperforming ITSx (69.9%) and ITSxpress v2 (41.4%).
  • Partial-Chain Impact: The fallback strategy recovered an additional 10,725 reads (19.6% of the dataset) that would have been discarded by a strict four-anchor requirement.
  • ITS2 Extraction: Achieved 93.1%, approaching ITSx's 97.1%.
• Boundary Accuracy:
  • On simulated data with known ground truth, all tools showed comparable accuracy for the Full ITS region (median absolute error 8–24 bp).
  • ITSxRust maintained high precision for subregions, filtering out reads with uncertain boundaries (unlike ITSx, which uses read edges as fallbacks, leading to higher boundary errors for truncated reads).
• Downstream Classification:
  • All extraction tools improved taxonomic classification accuracy (Genus/Species) compared to raw, untrimmed reads.
  • Differences between tools were negligible (<2%), indicating that the choice of extractor should be driven by speed and yield rather than classification fidelity.

5. Significance

ITSxRust addresses a critical gap in fungal metabarcoding pipelines for long-read sequencing. By combining Rust-native performance with a robust partial-chain recovery strategy, it maximizes data utility from expensive long-read runs, recovering reads that previous tools would discard.

The introduction of structured failure diagnostics is a major operational advancement, transforming extraction failures from opaque "black box" events into actionable insights for experimental design (e.g., identifying primer misplacement). While ITSxRust currently has a limitation regarding ITS1 extraction on reads lacking the SSU flank (unlike ITSx), this is a configurable trade-off for higher boundary confidence. The tool is available via GitHub, Bioconda, and Docker, facilitating immediate adoption in high-throughput fungal community studies.