DiaReport: Reproducible Workflow for Differential Expression Analysis and Interactive Reporting in DIA-based Proteomics

DiaReport is an open-source R package that streamlines reproducible differential expression analysis and generates interactive HTML reports for DIA-based proteomics by integrating DIA-NN outputs with MSqRob, QFeatures, and Quarto.

The Gist

Imagine you are a detective trying to solve a mystery inside a bustling city. In the world of biology, this "city" is a cell, and the "suspects" are thousands of tiny proteins that tell us how the body is working.

For a long time, looking for these proteins was like trying to catch specific birds in a flock by only looking at the ones that happen to fly into your net first. This method (called DDA) was hit-or-miss; you might catch the same few birds over and over while missing the rare, important ones.

Then, scientists invented a better way called DIA (Data-Independent Acquisition). Instead of picking birds, DIA takes a wide-angle photo of the entire flock, capturing every single bird, no matter how small or rare. It's a massive amount of data—like having a high-definition video of the whole city.

The Problem:
Now that we have this amazing, high-definition video, we have a new problem: How do we watch it?
The raw data is huge, messy, and hard to read. Scientists need to:

1. Clean up the video (remove static and bad frames).
2. Count the birds accurately.
3. Compare two different cities (e.g., a healthy city vs. a sick city) to see which birds are more common in one than the other.
4. Write a report that explains all this to other detectives.

Until now, doing this required a detective to be a master of three different languages (coding, statistics, and data visualization), switching between different tools, and manually stitching the results together. It was slow, prone to errors, and hard to repeat exactly.

The Solution: DiaReport
The authors of this paper have built a tool called DiaReport. Think of DiaReport as an automated "Smart City Dashboard" for protein detectives.

Here is how it works, using simple analogies:

1. The One-Click Magic Button

Instead of writing a complex script to clean the data, run the math, and make charts, you just feed DiaReport two things:

• The raw "video" from the machine (DIA-NN output).
• A simple list telling it which samples belong to which group (the "Experiment Design File").

DiaReport then acts like a conductor in an orchestra. It tells all the different instruments (statistical tools) when to play, ensuring they work in perfect harmony. It handles the cleaning, the counting, and the math in one smooth motion.

2. The "Filter" (Cleaning the Mess)

Imagine your city video has some blurry frames or birds that aren't actually part of the story (contaminants). DiaReport has a smart sieve.

• It can be set to keep only the birds that appear in every group (strict).
• Or, it can keep birds that appear in at least one group (lenient).
  This ensures you aren't making decisions based on blurry data or accidental noise.

3. The "Storyteller" (The Interactive Report)

This is the coolest part. In the past, scientists had to take screenshots of their graphs and paste them into a Word document. If they made a mistake, they had to start over.

DiaReport automatically builds a living, breathing website (an interactive HTML report).

• It's like a video game menu: You can click on a graph, zoom in, hover over a protein to see its name, and search for specific suspects.
• It's self-contained: You don't need a special server to view it. You just open the file, and the story is there.
• It has different "Costumes": The tool comes with different templates. If you are studying general proteins, it wears a "General Detective" suit. If you are studying Extracellular Vesicles (tiny bubbles cells use to talk to each other), it puts on a specialized "EV Detective" hat that looks for specific markers, just like a specialized police unit.

4. The "Time Machine" (Reproducibility)

Science is all about being able to repeat an experiment and get the same result.
DiaReport saves a recipe book (a configuration file) alongside the results. If another scientist wants to check your work, they can look at the recipe, see exactly how you cleaned the data and ran the math, and get the exact same answer. It removes the "black box" mystery of how the results were made.

Real-World Test: The "Bovine" Mystery

The authors tested this tool on a real case involving Extracellular Vesicles (EVs). They compared two ways of collecting these bubbles:

1. Method A (Ultracentrifugation): Like using a giant, slow-spinning centrifuge.
2. Method B (Ultrafiltration): Like using a high-tech coffee filter.

Using DiaReport, they quickly discovered that Method B was much cleaner. The tool's "dashboard" showed that Method A had accidentally picked up a lot of cow proteins (contaminants from the lab environment), while Method B was almost pure. Without this automated tool, spotting that subtle difference in the massive data would have taken weeks of manual work.

The Bottom Line

DiaReport is a bridge. It connects the complex, high-tech world of mass spectrometry data to the human need for clear, interactive, and trustworthy stories. It turns a mountain of raw numbers into a clear, clickable map that helps scientists understand what their cells are telling them, faster and with fewer headaches.

It's not just a tool; it's a translator that turns "computer code" into "biological insight."

Technical Summary

1. Problem Statement

While Data-Independent Acquisition (DIA) has become the standard for mass spectrometry-based proteomics due to its robustness and depth, the field lacks integrated, reproducible workflows for differential expression (DE) analysis and result reporting.

• Fragmented Tooling: Existing solutions often require users to switch between command-line tools (e.g., DIA-NN for quantification) and separate R packages (e.g., MSqRob, MSstats) for statistical modeling.
• Reporting Gaps: While tools exist for quality control (QC), few provide high-quality, interactive HTML reports specifically for DE analysis that link visualizations directly to the analysis version.
• Reproducibility Issues: Many current workflows rely on ad-hoc scripts and loosely structured objects, making it difficult to reproduce analyses, share results, or ensure interoperability with downstream tools.
• Accessibility: Complex pipelines often require significant programming expertise, hindering adoption in research groups with limited IT resources.

2. Methodology

DiaReport is an open-source R package designed to bridge these gaps by integrating data processing, statistical modeling, and interactive reporting into a single, streamlined workflow.

• Core Architecture:

  • Input: Accepts DIA-NN output (Parquet or TSV) and an Experiment Design File (EDF) containing sample metadata.
  • Data Processing Pipeline:
    1. Filtering: Applies precursor-level q-value filtering ($q \le 0.01$) and contaminant removal.
    2. Completeness Filtering: Implements three user-defined strategies based on a threshold $k$: per-group, across-groups, and global completeness.
    3. Transformation & Normalization: Performs log2-transformation and user-selectable normalization.
    4. Summarization: Aggregates precursor intensities to the protein level using robust methods (e.g., median polish).
  • Statistical Modeling: Utilizes MSqRob (paired with QFeatures) to perform DE analysis. This allows for flexible experimental designs (e.g., factorial models, blocking factors) without relying on imputation, as MSqRob explicitly models missing values.
  • Reporting Engine: Generates self-contained, interactive HTML reports using Quarto templates. Visualizations are powered by Plotly and dynamic tables by DataTables.

• Reporting Templates:

  • Base Template: Focuses on QC and standard DE analysis (volcano plots, heatmaps, PCA).
  • Partial Template: Adds "absent-from-DE" analysis to identify features detected only in specific conditions.
  • EV Template: Tailored for Extracellular Vesicle (EV) research, including specific marker panels (e.g., CD63, CD81) and contaminant interference analysis.

• Reproducibility Features:

  • All analysis parameters are saved in a YAML configuration file.
  • Results are stored in a structured directory containing RDS objects (serialized QFeatures), CSV top-tables, and PDF plots.
  • The workflow is platform-independent and does not require a running server environment (unlike Shiny apps).

3. Key Contributions

• End-to-End Integration: DiaReport is the first tool to seamlessly integrate DIA-NN quantification, MSqRob statistical modeling, and Quarto-based reporting into a single R function.
• Modular Reporting: The use of Quarto templates allows for domain-specific customization (e.g., the EV template) while maintaining a consistent core architecture.
• Enhanced Reproducibility: By generating structured outputs (YAML, QFeatures, CSV, PDF) and portable HTML reports, the tool ensures that analyses can be exactly reproduced and shared without dependency on active servers.
• Accessibility: The package simplifies complex statistical workflows for users with limited programming expertise, facilitating deployment in diverse research groups.

4. Results

The authors validated DiaReport using two distinct datasets:

• UPS2/Yeast Benchmark:

  • Used Universal Protein Standard 2 (UPS2) spiked into a yeast background.
  • Outcome: The tool accurately recovered expected protein fold-changes. Interactive PCA plots showed distinct clustering by spike-in levels, and UpSet plots confirmed high identification overlap across comparisons.

• Extracellular Vesicle (EV) Case Study:

  • Compared two EV enrichment protocols: Ultracentrifugation (UC) vs. 96-well Ultrafiltration (UF96).
  • Key Findings:
    • Contaminant Reduction: UF96 samples showed significantly lower abundance of bovine contaminants compared to UC.
    • Sample Consistency: UF96 replicates formed a tight cluster in PCA, whereas UC replicates showed higher variability and outliers.
    • Biological Insights: The interactive volcano plot revealed that transmembrane proteins were more abundant in UF96 samples. The "absent-from-DE" section successfully identified proteins unique to specific isolation methods.
  • Performance: For typical clinical cohorts (50–200 samples), the full workflow (analysis + report generation) completes within practical runtimes on a standard laptop (Supplementary Table S1).

5. Significance

DiaReport addresses a critical bottleneck in modern proteomics by transforming raw DIA-NN data into biologically interpretable, interactive, and reproducible reports.

• Standardization: It promotes the adoption of standardized, transparent workflows (FAIR principles) in the proteomics community.
• Efficiency: It eliminates the need for manual script writing and switching between disparate tools, significantly reducing the time from data acquisition to biological insight.
• Scalability: The modular design allows for future extensions to other quantitative workflows and downstream analyses (e.g., gene set enrichment), making it a foundational tool for the next generation of DIA-based proteomic studies.

Availability: The package is open-source (MIT license) and available at https://github.com/Gevaert-Lab/diareport. Interactive reports and supplementary data are archived on Zenodo.