deluxpore: a Nextflow pipeline for demultiplexing Illumina dual-indexed Nanopore libraries

The paper introduces **deluxpore**, a Nextflow pipeline that enables accurate demultiplexing of Nanopore reads from Illumina dual-indexed libraries by addressing challenges like residual adapters and high error rates, thereby facilitating reliable hybrid target-capture metagenomics workflows.

The Gist

Imagine you are running a massive, high-speed mail sorting facility.

The Problem: The "Noisy" Mail

In the world of DNA sequencing, scientists often want to find very rare, specific "letters" (genes) hidden inside a huge pile of "junk mail" (the rest of the genome). To do this, they use a technique called Target Capture, which is like using a magnet to pull out only the specific letters they care about.

Usually, this magnet works best with short, crisp letters (Illumina sequencing). But scientists also want to use Nanopore sequencing, which is like a super-fast, long-distance courier. It can read entire, long letters in one go, which is great for understanding the full story. However, Nanopore couriers are a bit "noisy" and make mistakes (typos) much more often than the standard short-read couriers.

The Conflict:

1. The Magnet: Needs letters to have specific, clean "address labels" (indexes) to know which pile they belong to.
2. The Courier: Delivers the letters, but the address labels are often smudged, torn, or buried under extra wrapping paper (adapter fragments) because the courier is so fast and error-prone.

Standard sorting software is like a robot that expects perfect, crisp labels. If the label is smudged or the wrapping paper is in the way, the robot throws the letter in the "trash" (unassigned data). This meant scientists couldn't use the powerful long-read couriers for their targeted magnet experiments.

The Solution: deluxpore (The Smart Sorter)

The authors of this paper built a new tool called deluxpore. Think of it as a super-smart, patient human sorter who can handle messy mail.

Instead of just looking for a perfect match, deluxpore uses two clever tricks:

1. The "Fuzzy Match" (BLAST & Levenshtein Distance): If a label says "New York" but the courier wrote "Nw Yrok," a normal robot says, "Wrong!" and throws it away. deluxpore says, "Ah, that's close enough to New York. Let's count the typos and figure it out." It can handle the smudges and missing letters.
2. The "Wrapper" Awareness: It knows that sometimes the address label is stuck under a piece of tape (the adapter). It knows how to look under the tape to find the real address.

The Experiment: Testing the Sorter

The team tested this new sorter with two scenarios:

Scenario A: The "Combinatorial" Pile (96 Samples)
Imagine they tried to sort 96 different people's mail by giving them labels that were combinations of two colors (e.g., Red-Blue, Red-Green).

• The Result: It was chaotic. If the "Red" part of the label was smudged, the sorter couldn't tell if it was "Red-Blue" or "Red-Green." Even with high-quality mail, they only recovered about 46% of the letters. It was too confusing.

Scenario B: The "Unique" Pile (8 Samples)
They realized that if they gave each person a completely unique, one-of-a-kind label (no shared colors), the sorter could work much better. Even if one part of the label was smudged, the other unique part was enough to identify the owner.

• The Result: With high-quality mail, they recovered 91.7% of the letters. With the best quality, they hit nearly 98%.

The Big Takeaway

The paper teaches us two main lessons for the future of DNA research:

1. Quality Matters: You need the mail to be reasonably clean (a quality score of Q20 or higher). If the letters are too messy, even the smartest sorter can't help.
2. Design Matters More: It's better to use a simple, unique labeling system (one unique label per person) rather than a complex, shared system. By avoiding "confusing" label pairs, scientists can get almost perfect results.

In short: deluxpore is the bridge that finally lets scientists use the fast, long-read Nanopore couriers for their targeted DNA experiments, provided they use a smart labeling strategy and decent quality mail. It turns a "broken" workflow into a reliable, automated system.

Technical Summary

1. Problem Statement

The integration of target capture metagenomics (selective enrichment of specific genomic regions) with long-read sequencing (Oxford Nanopore Technologies, ONT) offers a powerful method for characterizing rare microbial taxa and full-length functional genes. However, a critical technical bottleneck exists:

• Incompatibility: Standard ONT barcoding kits are incompatible with established target capture workflows, which are optimized for short-read Illumina platforms.
• The Hybrid Workaround: A viable solution involves preparing libraries with Illumina dual-index adapters (e.g., NEBNext or Nextera), performing target capture, and then converting the enriched pool into an ONT-compatible library.
• The Software Gap: Standard Illumina demultiplexing tools fail on ONT data because:
  1. High Error Rates: Nanopore raw reads have significantly higher error rates (5–15%) compared to Illumina (0.1–1%), obscuring short index sequences.
  2. Positional Variability: Long reads exhibit variable positioning of adapter fragments, which standard algorithms cannot handle.
  3. Residual Adapters: Indexes are often embedded within residual adapter fragments rather than being cleanly separated.

2. Methodology: The deluxpore Pipeline

The authors developed deluxpore, a Nextflow pipeline designed to robustly demultiplex Nanopore reads derived from Illumina dual-indexed libraries.

• Workflow Stages:

  1. Preprocessing: Trimming of Nanopore adapters and quality filtering (using Porechop and Chopper). Reads are converted to FASTA format.
  2. Adapter Identification: Uses BLAST alignment against a custom database containing only the specific oligo sequences used in the experimental design. This reduces computational overhead and spurious mappings.
  3. Index Extraction & Matching:
     • Extracts unique index sequences (8 nt for NEBNext, 10 nt for Nextera) from fixed positions flanking the mapped adapter regions.
     • Compares extracted sequences against a reference library using Levenshtein distance (edit distance) to find the best-matching i5 and i7 index pairs, accommodating sequencing errors.
  4. Sample Assignment Logic:
     • Hierarchical Decision: Prioritizes the lowest Levenshtein distance, then alignment position (i5 near start, i7 near end).
     • Dual vs. Single Index Assignment:
       • Dual-index assignment: Required when indexes are shared across multiple samples (combinatorial design). Both i5 and i7 must match.
       • Single-index assignment: Enabled when each index uniquely identifies a sample. If one index fails, the read can still be assigned based on the other, increasing recovery rates.
     • Ambiguity Handling: Reads with identical distance/position values mapping to different valid samples are flagged as ambiguous and excluded to prevent misclassification.

• Implementation: Built with Nextflow, Python, and Bash. It supports parallel processing of read chunks, making it scalable for local workstations and HPC environments.

3. Key Contributions

• Specialized Software: The first tool capable of demultiplexing Nanopore reads from Illumina dual-indexed libraries, bridging the gap between target capture protocols and long-read sequencing.
• Algorithmic Innovation: Utilizes BLAST for adapter detection and Levenshtein distance for error-tolerant index matching, specifically addressing the high-error nature of Nanopore data.
• Optimized Experimental Design: Identified specific "high-crosstalk" index pairs within the NEBNext Primer Set A and provided an optimized 8-sample configuration to minimize cross-assignment errors.
• Open Source: The pipeline, documentation, and benchmarking workflows are publicly available under the GNU GPL v3.0.

4. Benchmarking Results

The authors validated the pipeline using two simulated datasets (96-sample combinatorial and 8-sample unique) across 18 replicates with varying quality scores (Q10, Q20, Q25, Q30).

• Impact of Index Design:

  • Combinatorial (96-sample): Required both indexes to be identified. At Q20, sample recovery was only 46.1%.
  • Unique (8-sample): Allowed single-index assignment. At Q20, sample recovery jumped to 91.7%.
  • Conclusion: Unique index-to-sample designs significantly outperform combinatorial designs, especially at moderate quality levels.

• Quality Thresholds:

  • Accurate demultiplexing requires a minimum Q20 data quality.
  • The optimized 8-sample design (excluding high-crosstalk pairs like i704-i706) achieved >98% accuracy at Q20.
  • Combinatorial designs require exceptional quality (Q30) to match the performance of unique designs at Q20.

• Error Analysis:

  • Crosstalk: Specific index pairs (e.g., i7010-i702, i704-i706) showed significant confusion. Removing these pairs in the 8-sample design reduced cross-assignment to negligible levels (≤0.7%).
  • Failure Mode: Demultiplexing failures were primarily driven by sequence truncation (partial reads) rather than base-calling errors within the index regions themselves.

5. Significance

• Enables Hybrid Workflows: deluxpore makes it feasible to combine target capture enrichment with long-read sequencing, a combination previously hindered by a lack of software.
• Scalability: The pipeline supports high-throughput processing, enabling large-scale studies of rare microbial taxa and functional genes.
• Best Practices: The study provides concrete guidelines for researchers: use unique dual-index designs and ensure Q20+ quality to achieve reliable results.
• Cost-Effectiveness: By enabling the use of established, cost-effective Illumina capture kits with ONT sequencing, it lowers the barrier for deep functional characterization of complex microbiomes.