Pairing Data Independent Acquisition and High-Resolution Full Scan for Fast Urinary Tract Infection Diagnosis

This study presents a rapid, culture-free workflow for diagnosing urinary tract infections by combining high-resolution data-independent acquisition to create a pathogen-specific peptide reference panel with cost-effective MS1-only screening and machine learning, achieving high accuracy in identifying uropathogens from both synthetic and clinical urine samples.

The Gist

Imagine you are a detective trying to solve a crime, but the culprit (a bacteria causing a urinary tract infection) is hiding in a massive, chaotic crowd of innocent people (your urine).

Traditionally, to catch the culprit, you have to wait for them to grow a "clone army" in a lab dish (a process called culture) until they are big enough to be seen. This takes days. In the meantime, doctors often guess which antibiotic to use, which can be ineffective and help bacteria become resistant to medicine.

This paper presents a new, super-fast detective method that skips the waiting game entirely. Here is how it works, broken down into simple concepts:

1. The Two-Tool Strategy: The "Master Blueprint" and the "Quick Snapshot"

The researchers used two different types of high-tech cameras (mass spectrometers) to solve the case.

• The Master Blueprint (The Orbitrap Astral): Think of this as a super-powerful, expensive camera that takes incredibly detailed, slow-motion photos of the bacteria. It breaks the bacteria down into tiny pieces (peptides) and identifies every single one with perfect clarity. The researchers used this to create a "Wanted Poster" for eight common types of bad bacteria. This poster lists the unique "fingerprints" (specific chemical signatures) of each criminal.
• The Quick Snapshot (The Orbitrap Exploris 480): This is a more affordable, faster camera that is common in hospitals. It can't take the slow-motion, detailed photos, but it can take a super-fast snapshot of the whole crowd in just a few minutes. It sees the crowd as a blur of shapes and colors (mass and time), but it can't easily tell who is who just by looking.

The Innovation: The team figured out how to use the Master Blueprint to teach the Quick Snapshot how to recognize the criminals. They trained a computer to look at the fast, blurry snapshot and say, "Ah, I see that specific shape and color pattern from the Wanted Poster! That must be E. coli!"

2. The "Face Recognition" AI

Once they had the "Wanted Posters" (the list of unique bacterial fingerprints), they didn't just look at them manually. They taught a Machine Learning AI (a computer brain) to be the detective.

• Training: They showed the AI thousands of "snapshots" of urine samples containing known bacteria. The AI learned to ignore the noise and focus on the specific patterns that matched the Wanted Posters.
• The Test: They then gave the AI real patient urine samples it had never seen before. The AI looked at the fast snapshot, matched the patterns to its memory of the Wanted Posters, and shouted out the name of the bacteria.

3. The Results: Speed and Accuracy

• Speed: Instead of waiting 24–72 hours for bacteria to grow, this method gives a result in minutes. It can process hundreds of samples a day, like a high-speed conveyor belt.
• Accuracy: The AI was incredibly good at its job. On test samples, it was right about 92% of the time. On real patient samples, it was right about 77% of the time.
• The "Edge Cases": When the bacteria were very weak (low concentration), the AI sometimes got confused and said "No one here" (a false negative). However, even when it was unsure, it often gave a "second guess" that was actually the right bacteria, suggesting the method is sensitive enough to catch even faint signals if we look closely at the confidence scores.

4. Why This Matters

Think of this like upgrading from waiting for a letter to arrive by mail (the old culture method) to getting an instant text message (this new method).

• Better Treatment: Doctors can prescribe the exact right antibiotic immediately, rather than guessing.
• Fighting Superbugs: By using the right drug immediately, we stop bacteria from learning how to resist our medicines.
• Accessibility: The "Quick Snapshot" camera is cheaper and easier to find in hospitals than the "Master Blueprint" camera. This means the technology can be used in regular clinics, not just fancy research labs.

The Bottom Line

The researchers built a bridge between a high-end research tool and a standard hospital tool. By using the high-end tool to create a "cheat sheet" of bacterial fingerprints, they taught the standard tool to identify infections instantly. It's a fast, culture-free way to catch urinary tract infections, helping doctors treat patients faster and smarter.

Technical Summary

1. Problem Statement

• Clinical Need: Urinary Tract Infections (UTIs) are a massive global health burden, yet current diagnostic methods rely on bacterial culture, which takes 24–72 hours. This delay leads to empirical antibiotic use, contributing to the rise of antimicrobial resistance (AMR).
• Technological Limitations: While Liquid Chromatography-Tandem Mass Spectrometry (LC-MS/MS) offers rapid, culture-free identification, standard workflows are resource-intensive and slow.
  • High-end instruments (e.g., Orbitrap Astral): Offer high-speed, deep peptide profiling via Data-Independent Acquisition (DIA) but are too expensive and complex for routine clinical use.
  • Routine clinical instruments (e.g., Orbitrap Exploris 480): Are cost-effective and high-throughput but lack the speed to perform full MS/MS fragmentation on all target peptides within the short run times (e.g., 5 minutes) required for clinical labs (300 samples/day).
  • MS1-only limitations: Running these routine instruments in MS1-only (Full Scan) mode is fast but suffers from high noise and peptide misidentification without fragmentation data.

2. Methodology

The authors propose a hybrid workflow that leverages the strengths of two different mass spectrometers to create a rapid, culture-free diagnostic pipeline.

• Hybrid Data Acquisition Strategy:

  • Reference Generation (DIA): High-resolution Data-Independent Acquisition (DIA) is performed on an Orbitrap Astral instrument. This generates a deep, high-confidence reference panel of pathogen-specific peptides (precursors) with precise m/z, retention time (RT), and isotopic patterns.
  • Routine Analysis (MS1-only): Clinical and validation samples are analyzed on a more accessible Orbitrap Exploris 480 in Full Scan (MS1-only) mode. This allows for very fast run times (approx. 5 minutes per sample).
  • Data Pairing: An R-based pipeline matches the MS1 features from the Exploris 480 against the DIA-derived reference library using strict criteria: mass tolerance (±3 ppm), retention time window (±5 seconds), and isotopic cluster presence (M, M+1, M+2).

• Machine Learning Framework:

  • Training Data: The model was trained on high-concentration inoculated urine samples ($10^6$ CFU/mL) and pure bacterial cultures of eight common uropathogens (E. coli, E. faecalis, K. pneumoniae, P. aeruginosa, S. agalactiae, S. aureus, S. saprophyticus, P. mirabilis) and negative controls.
  • Algorithm: Nine One-vs-All Random Forest classifiers were trained. Each classifier distinguishes one specific class (pathogen or blank) from all others.
  • Feature Selection: From an initial set of 8,218 DIA-identified precursors, 4,281 were consistently detected across datasets. Univariate filtering (InfoGain) reduced this to 2,612 features for model training.
  • Prediction Logic: A "winner-takes-all" approach selects the class with the highest probability score. Confidence is assessed via the top score and the margin between the top two scores.

• Software Implementation:

  • Data processing and modeling were performed in R.
  • An RShiny application was developed to automate mzML preprocessing, feature extraction, and classification, supporting parallel processing for large datasets.

3. Key Contributions

• Novel Workflow: Demonstrates a proof-of-concept for pairing deep DIA profiling on research-grade instruments with rapid MS1-only acquisition on routine clinical instruments.
• Culture-Free Speed: Achieves pathogen identification in approximately 5 minutes per sample, bypassing the 24–72 hour culture delay.
• Scalability: The workflow is designed for high-throughput clinical environments (300 samples/day) using cost-effective instrumentation.
• Interpretability: Introduces a confidence scoring system (probability and margin) to help clinicians distinguish between high-confidence diagnoses and ambiguous cases (e.g., low bacterial load or unknown pathogens).
• Open Resources: Provides the code (GitHub) and raw data (ProteomeXchange) for reproducibility.

4. Results

• Biomarker Discovery: The DIA analysis on the Orbitrap Astral identified 8,218 candidate peptides. After filtering and matching to MS1 data, 4,281 precursors were retained as a robust feature set.
• Model Performance (Internal Validation):
  • On held-out test data (70/30 split of inoculated samples), the model achieved a Matthews Correlation Coefficient (MCC) of 0.924.
  • Training MCCs were near perfect (1.0 for most classes).
• Model Performance (Clinical Validation):
  • On an independent set of 127 clinical patient samples, the model achieved an overall MCC of 0.77.
  • Correct classification rates exceeded 83% for six of the eight target pathogens.
  • Limitation: Streptococcus agalactiae showed lower performance (42% correct) in patient samples, attributed to low bacterial concentrations in clinical specimens falling below the detection threshold of the MS1-only workflow.
• Edge Cases:
  • Low Concentration ($10^5$ CFU/mL): Often misclassified as "blank" (negative), but the secondary prediction scores frequently indicated the correct species, suggesting the model retains partial discriminatory power.
  • Unknown Pathogens: When presented with non-target species, the model tended to assign them to the most taxonomically similar known pathogen (e.g., Serratia marcescens predicted as K. pneumoniae), demonstrating that the model learns shared peptide signatures.

5. Significance

• Clinical Impact: This method offers a viable path toward same-day, targeted antibiotic therapy, directly addressing the crisis of antimicrobial resistance by reducing the reliance on empirical treatment.
• Economic Viability: By utilizing the Orbitrap Exploris 480 (a standard clinical instrument) rather than requiring expensive, specialized DIA-only workflows, the method is scalable for routine hospital laboratories.
• Future Directions: The study establishes a foundation for:
  • Expanding the pathogen library to include more species and strains.
  • Developing quantitative workflows to distinguish between contamination (low load) and true infection (high load).
  • Applying the "DIA-to-MS1 pairing" strategy to other infectious diseases and biomarker discovery contexts.

In conclusion, the paper successfully demonstrates that combining high-resolution reference libraries with rapid, low-cost MS1 acquisition and machine learning can overcome the speed and cost barriers of traditional proteomics, enabling rapid, culture-free UTI diagnosis.