DIANA: An integrated pipeline for analysis of long-read whole-genome sequencing data for molecular neuropathology.

The paper introduces DIANA, an automated pipeline that integrates long-read whole-genome sequencing data to simultaneously generate comprehensive molecular profiles—including methylation classification, copy-number variants, gene fusions, and small variants—for streamlined CNS tumor diagnosis and clinical decision-making.

The Gist

Imagine you are a detective trying to solve a very complex crime: a brain tumor. In the past, to get all the clues, you had to send evidence to five different labs. One lab would look at the DNA letters, another would check the chemical "sticky notes" (methylation) attached to the DNA, a third would look for missing pages in the book (deletions), and so on. It was expensive, slow, and the reports often didn't match up perfectly.

DIANA is like a super-smart, all-in-one detective agency that solves the whole case in a single day using just one piece of evidence: a long, continuous strand of DNA.

Here is how the paper explains this new tool, broken down into simple concepts:

1. The Problem: The "Jigsaw Puzzle" Nightmare

Currently, diagnosing brain tumors is like trying to solve a jigsaw puzzle where the pieces are scattered across five different tables. Doctors have to use different machines to find:

• Small typos in the DNA (mutations).
• Missing or extra pages in the DNA book (copy number changes).
• Big structural rearrangements (fusions).
• Chemical tags on the DNA that tell the cell how to behave (methylation).

Doing all this separately is a logistical and financial headache.

2. The Solution: The "Swiss Army Knife" Sequencer

New technology (called Long-Read Sequencing) is like a camera that can take a photo of the entire DNA strand in one go, rather than snapping tiny, disconnected pictures. It can see the typos, the missing pages, the big rearrangements, and the chemical tags all at the same time.

But here's the catch: The camera takes a picture, but it doesn't tell you what the picture means. You still need an expert to interpret it.

3. Enter DIANA: The "Auto-Pilot" Interpreter

DIANA (Diagnostic Integrated Analytics of Neoplastic Alterations) is the software that acts as that expert. It is a fully automated pipeline that takes the raw photo from the camera and turns it into a clear, easy-to-read medical report.

Think of DIANA as a four-step assembly line in a factory:

• Step 1: The Merging Station (mergebam)
  Imagine the DNA camera takes photos in chunks throughout the day. DIANA takes all these scattered photo strips and glues them together into one perfect, long scroll. It also highlights the specific pages the doctor cares about most.
• Step 2: The Detective Work (epi2me)
  This is the heavy lifting. DIANA uses different "detectives" to look for specific clues simultaneously:
  • One detective checks the chemical tags (methylation) to guess what type of tumor it is.
  • Another looks for missing pages or extra copies (Copy Number Variations).
  • A third hunts for typos (mutations) and big structural breaks.
• Step 3: The Translator (annotation)
  Now that the clues are found, DIANA translates them into human language. It asks: "Is this typo dangerous?" "Does this missing page explain the tumor?" It cross-references the findings with massive databases of known cancer secrets to see if the clues match a known criminal profile.
• Step 4: The Final Report (report generation)
  Finally, DIANA writes a summary. It gives the doctor a one-page "Executive Summary" (like a cheat sheet) with the most critical answers, followed by a detailed "Extended Report" with all the evidence, charts, and graphs if they want to dig deeper.

4. Why is this a Big Deal?

• Speed & Simplicity: Instead of waiting weeks for results from five different labs, DIANA can process the data in about 82 minutes on a standard computer.
• One Source of Truth: Because it looks at everything at once, the results are consistent. The "methylation" report matches the "mutation" report because they came from the same analysis.
• Open Source: The creators made the "blueprints" for DIANA free for anyone to use. It's like giving every hospital the recipe for this super-detective, so they can build their own version without paying a fortune.
• No PhD Required: The tool is designed so that a hospital technician can run it without needing to be a computer genius. It handles the complex math in the background.

The Bottom Line

DIANA is a game-changer for brain tumor diagnosis. It takes a complex, fragmented process and turns it into a single, automated, and clear workflow. It's like upgrading from a team of five people arguing over different maps to a single GPS that instantly shows you the exact location of the tumor and the best path to treat it.

Note: The paper mentions that while this tool is powerful and ready for research, it is currently a "research tool" and doctors must still use their own judgment before making final clinical decisions.

Technical Summary

1. Problem Statement

The accurate diagnosis and classification of Central Nervous System (CNS) tumors require a comprehensive genomic profile that includes:

• DNA methylation patterns (essential for WHO 2021 classification).
• Copy Number Variations (CNVs).
• Structural Variants (SVs) and gene fusions.
• Small variants (SNVs and indels).
• MGMT promoter methylation status (critical for treatment response prediction).

Current Limitations:

• Fragmented Workflows: Existing diagnostic workflows rely on multiple separate platforms (NGS, methylation arrays, pyro-sequencing, FISH), creating significant logistical, financial, and time burdens.
• Integration Complexity: While long-read sequencing (Oxford Nanopore and PacBio) can capture all these data types in a single assay, integrating diverse analytical tools into a unified, automated, and reproducible pipeline for clinical use has remained a challenge.
• Lack of Automation: There was no publicly available, end-to-end automated pipeline specifically designed for tumor diagnostics using long-read WGS data.

2. Methodology

DIANA (Diagnostic Integrated Analytics of Neoplastic Alterations) is an open-source, containerized pipeline built on Nextflow (DSL 2). It is designed to process aligned long-read BAM files from ONT or PacBio platforms into a unified diagnostic report.

The pipeline consists of four modular stages:

A. Module 1: mergebam (Data Preprocessing)

• Function: Handles raw BAM files generated during sequencing runs (e.g., hourly splits from MinKNOW).
• Mechanism: Merges split BAM files based on sample and flow cell identifiers to prevent sample mix-ups.
• Output: A single merged, indexed BAM file and a Region-of-Interest (ROI) extracted BAM file (filtered for SNV calling).

B. Module 2: epi2me (Primary Genomic & Epigenetic Analysis)

This module runs parallel analyses on the merged data:

• Methylation Quantification: Uses Modkit to extract base modification tags (5mC and 5hmC) genome-wide, producing per-CpG quantification.
• Structural Variant (SV) Calling: Uses Sniffles2 to detect insertions, deletions, inversions, and translocations.
• Copy Number Profiling: Uses QDNAseq (with GC-content and mappability correction) to generate genome-wide CNV segments and bin-level data.
• Small Variant Calling: Uses Clair3 (germline) and ClairS-TO (somatic, tumor-only) on the ROI BAM file to detect SNVs and indels.
• Quality Control: Computes read statistics (coverage, length distribution) using cramino.

C. Module 3: annotation (Interpretation & Prioritization)

This module interprets raw outputs for clinical relevance:

• MGMT Status: Quantifies methylation at the MGMT promoter CpG island to determine methylated vs. unmethylated status.
• Tumor Classification: Runs two independent neural network classifiers (CrossNN and Sturgeon) using reference datasets (Capper et al., 2018) to assign WHO 2021 tumor classes and families with probability scores. Includes UMAP visualization for sample positioning.
• SV Annotation: Uses Svanna to cross-reference SVs with Human Phenotype Ontology (HPO) and extract gene fusion events.
• CNV Visualization: Generates annotated CNV plots (using R/ggplot2) highlighting gains/losses in specific genes (e.g., EGFR, CDKN2A/B).
• Tumor Content Estimation: Uses ACE to estimate tumor cell content from CNV data.
• Variant Filtering: Merges and filters VCFs against ClinVar and COSMIC databases to produce a ranked variant table.

D. Module 4: report generation

• Consolidates all results into a comprehensive PDF report via R Markdown.
• Structure:
  • Executive Summary: One-page clinical overview (coverage, tumor content, methylation class, key CNVs/SVs, MGMT status).
  • Extended Report: Detailed statistics, full classification scores, UMAP plots, chromosome-specific CNV profiles, and full variant tables.
• Supplementary Outputs: Interactive HTML files (Svanna, UMAP) and formatted BED files for compatibility with other platforms.

Execution Modes:
The pipeline supports flexible execution, including "Smart sample monitoring" (auto-triggering upon sequencing completion), partial runs (e.g., merging only, or re-running annotation only), and full end-to-end processing.

3. Key Contributions

1. First End-to-End Long-Read Diagnostic Pipeline: DIANA is the first publicly available pipeline to automate the entire diagnostic workflow for CNS tumors using long-read WGS, from raw BAM to clinical report.
2. Unified Multi-Omics Approach: It successfully integrates DNA methylation, CNV, SV, and SNV analysis into a single workflow, eliminating the need for separate assays.
3. Reproducibility and Portability: Built with Nextflow and fully containerized (Docker/Apptainer), ensuring results are reproducible across local workstations and High-Performance Computing (HPC) clusters.
4. Clinical Utility: Generates a human-readable, structured report specifically tailored for neuropathologists, highlighting WHO 2021 classifications and actionable biomarkers (e.g., MGMT, IDH, TERT).
5. Open Source: Freely available under an MIT license with comprehensive documentation on GitHub.

4. Results & Performance

• Benchmarking: Tested on a glioblastoma sample with 28.78× coverage.
• Runtime: The full pipeline completed in approximately 82 minutes on a single workstation (190 GB RAM, 48 CPU cores) using parallel processing.
  • Bottlenecks: Somatic SNV calling (ClairS-TO) took ~59 mins; BAM merging took ~19 mins.
  • Annotation: The interpretation module was lightweight (<4 mins).
• Resource Requirements: Minimum recommended configuration is 32 GB RAM, 8 CPU cores, and 500 GB disk space.
• Accuracy: The pipeline successfully identified:
  • Methylation class: Glioblastoma, IDH wildtype, subclass RTK II (Score: 0.571).
  • CNVs: EGFR amplification (Copy number 32), CDKN2A/B homozygous deletion, and chromosome 7 gain/10 loss.
  • SNVs: Pathogenic PIK3CA and TERT promoter mutations.
  • MGMT Status: Unmethylated (16.11% mean methylation).
• Robustness: All processes completed without out-of-memory errors.

5. Significance

• Clinical Transformation: DIANA addresses the logistical and financial barriers of CNS tumor diagnostics by replacing multiple fragmented tests with a single, automated long-read sequencing workflow.
• Standardization: It provides a standardized framework for capturing all diagnostically relevant genomic dimensions, reducing variability between labs.
• Accessibility: By running on standard workstations and requiring no prior bioinformatics expertise (due to automation), it democratizes access to advanced long-read genomic diagnostics.
• Future-Proofing: The modular design allows for the easy integration of new tools or reference datasets (e.g., updated WHO classifications) without rebuilding the entire pipeline.

In conclusion, DIANA represents a significant step forward in molecular neuropathology, demonstrating that long-read sequencing can be effectively translated into a practical, automated, and comprehensive clinical diagnostic tool for CNS tumors.