NovoTax: prokaryotic strain identification from mass spectrometry-based proteomics data

NovoTax is an end-to-end pipeline that enables strain-level identification of prokaryotic organisms directly from raw mass spectrometry-based proteomics data by combining de novo peptide sequencing with optimized genome database searching, thereby facilitating downstream proteome analysis without prior knowledge of sample composition.

The Gist

Imagine you walk into a bustling, chaotic kitchen and find a pile of half-eaten meals on the table. Your goal is to figure out exactly who cooked these dishes and what specific recipe they used, but you don't have the cook's name or the recipe book.

In the world of biology, scientists often face this exact problem. They have a sample of bacteria (the "meal") and they use a machine called a Mass Spectrometer to break the proteins down into tiny pieces (the "ingredients"). Traditionally, to identify the bacteria, scientists need to know exactly which bacteria they are looking for beforehand so they can match the ingredients to a known recipe book. If they guess the wrong bacteria, the whole analysis fails.

Enter "NovoTax": The Detective Chef.

This new software, called NovoTax, is like a brilliant detective who doesn't need a suspect list to solve the crime. Instead, it looks at the ingredients (the protein fragments) and figures out who the chef is from scratch.

Here is how it works, broken down into simple steps:

1. The Taste Test (De Novo Sequencing)

First, NovoTax looks at the raw data from the machine. It's like a chef tasting a sauce and saying, "Hmm, this tastes like garlic, lemon, and a hint of rosemary."

• The Magic: Instead of guessing, NovoTax uses advanced AI (like a super-smart taste bud) to reconstruct the exact "flavor profile" (peptide sequence) of the ingredients, even if it has never seen this specific dish before.

2. The Library Search (The Database)

Once it has the list of ingredients, it needs to find the cookbook. But the library is huge—it contains the recipes for every known bacteria on Earth (a database called GTDB with over 700,000 genomes). Searching through all of them at once would be like trying to read every book in the Library of Congress to find one sentence. It would take forever!

NovoTax's Shortcut:

• Step 1 (The Neighborhood): It first looks for the "neighborhood" (the Genus). It asks, "Is this a Streptomyces dish or an E. coli dish?" It narrows the search down from 700,000 books to just a few dozen.
• Step 2 (The Family): Once it knows the neighborhood, it looks at the specific family of recipes.
• Step 3 (The Exact Strain): Finally, it finds the exact "strain" (the specific version of the bacteria). It's like finding the exact edition of a cookbook that matches the chef's handwriting.

3. The "Wait, There's More!" Feature (Contaminants)

Sometimes, a sample isn't just one type of bacteria; it's a mix, or maybe there's a "stowaway" (a contaminant) hiding in the sample.

• The Analogy: Imagine you are trying to identify a pizza, but you find a slice of pineapple on it that doesn't belong.
• NovoTax's Trick: After it identifies the main bacteria, it looks at the leftover ingredients that didn't fit the first match. It says, "Okay, these extra bits don't belong to the main chef. Let's run the search again to see who else was cooking." This allows it to spot hidden contaminants that other tools miss.

Why Does This Matter?

• No Guessing Required: You don't need to know what you are looking for before you start. NovoTax figures it out for you.
• Strain-Level Precision: It doesn't just say "It's a dog." It says, "It's a Golden Retriever named Spot." This is crucial because different strains of the same bacteria can be harmless or deadly.
• Quality Control: It acts like a security guard. If you think you are studying a specific bacteria, NovoTax can tell you, "Actually, your sample is mostly full of a different bacteria, or there's a contaminant."
• Community Detective: It can also look at a complex soup of many different bacteria (like in soil or the human gut) and tell you which ones are the "big shots" (the most abundant).

The Bottom Line

NovoTax is a tool that turns a messy pile of protein data into a clear answer: "This sample contains this specific bacteria, and here is its exact genetic recipe." It saves scientists time, prevents them from studying the wrong bugs, and helps them discover new things in the microbial world without needing a crystal ball to guess what's inside the sample first.

Technical Summary

1. Problem Statement

Traditional mass spectrometry (MS)-based proteomics workflows rely heavily on prior knowledge of the sample's composition. To identify peptides, spectra must be matched against a pre-existing protein database containing the specific organisms present in the sample. This creates a circular dependency: researchers must know the species to analyze the proteins, yet proteomics is often used to determine the species or strain in clinical or environmental contexts.

Current solutions lack an end-to-end pipeline that starts with raw MS data and produces a sample-specific protein database ready for conventional search engines. While de novo peptide sequencing exists, and taxonomic inference tools exist, no integrated solution currently bridges raw data directly to strain-level identification without a reference database of the specific sample.

2. Methodology: The NovoTax Pipeline

NovoTax is a modular, end-to-end pipeline designed to identify prokaryotic strains directly from raw bottom-up proteomics data (both Data-Dependent Acquisition [DDA] and Data-Independent Acquisition [DIA]). It operates in three iterative steps:

A. De Novo Sequencing

• Input: Raw MS files.
• Tools:
  • XuanjiNovo: Used for DDA data.
  • Cascadia: Used for DIA data.
• Filtering: The pipeline generates a peptide list, retaining only sequences with a prediction confidence score > 0.8. Users can also input pre-existing peptide tables.

B. Database Peptide Matching (Iterative Search)

Instead of searching the entire Genome Taxonomy Database (GTDB) at once (which contains ~732,000 proteomes and 2.4 billion sequences), NovoTax employs a hierarchical, three-step search strategy using MMseqs2 to optimize speed and memory usage:

1. Genus Level: Searches against a reduced database of genus representative species (approx. 7,993 proteomes). If a genus representative is missing, it selects a species with >95% completeness and <1% contamination.
2. Family Level: Once a genus is identified, the search expands to all species representatives within that genus's family.
3. Strain Level: Finally, peptides are searched against all strains of the identified species.

• Chemical Handling: Isoleucine is converted to leucine in both the query peptides and the database to account for MS indistinguishability.

C. Taxonomy Assignment & Scoring

• Scoring Metric: A score is calculated for each peptide: $Score = \frac{\text{Alignment Bitscore}}{\text{Number of Proteomes Matched}}$. This penalizes peptides that match multiple proteomes (low information content).
• Selection: For each search step, the proteome with the highest sum of peptide scores is selected.
• Iterative Contaminant Removal: After a strain is identified, matched peptides are removed. The remaining unmatched peptides are re-searched to identify potential contaminating species.
• Stopping Criterion: The process halts if the genus score drops below 1,402 (an empirically determined threshold based on the median score of the second round in benchmark samples).

3. Key Contributions

• End-to-End Automation: NovoTax is the first pipeline to convert raw MS files directly into a strain-specific protein FASTA database, enabling downstream analysis with traditional search engines (e.g., MSFragger).
• Hierarchical Search Strategy: The tiered approach (Genus $\to$ Family $\to$ Strain) significantly reduces computational load when querying the massive GTDB.
• Contaminant Detection: The iterative removal of matched peptides allows the tool to identify not just the dominant organism, but also low-abundance contaminants within a sample.
• Open Source & Modular: The software is available as open-source Docker containers, making it accessible to non-experts and allowing users to plug in their own intermediate results.

4. Results & Validation

The authors validated NovoTax using public datasets from PRIDE (Lee et al., 2019; Schape et al., 2019; Wuyts et al., 2023).

• Single-Species Isolates:

  • Accuracy: Identified the reported species in 90.2% (46/51) of cases.
  • Discrepancy Resolution: In cases where NovoTax identified a different species within the same genus, traditional proteomics searches against a concatenated database (NovoTax strain + Reported strain) revealed that >30% of peptides uniquely matched the NovoTax strain, while <2% matched the reported strain. This suggests the original sample annotations were likely incorrect.
  • Strain-Level Precision: For 84.4% of cases with strain annotations, NovoTax identified a strain with >99.5% Average Nucleotide Identity (ANI) to the reported strain.
  • Contaminants: Successfully identified contaminating species in samples (e.g., Streptomyces venezuelae and Halomonas sp.), where re-analysis of unassigned peptides revealed a second species accounting for >39% of the peptide spectrum.

• Bacterial Communities:

  • Simple Community (8 species): Identified 5 of the 8 reported species. However, these 5 accounted for >99% of the peptides in a traditional search, indicating they were the dominant organisms.
  • Complex Community (96h time point): Identified 11 species, 9 of which were among the most abundant species reported in the study.
  • DIA Performance: While effective, DIA datasets showed a higher fraction of unassigned peptides compared to DDA, suggesting DIA de novo sequencing is currently less precise.

5. Significance

NovoTax addresses a critical bottleneck in microbial proteomics by removing the requirement for prior taxonomic knowledge. Its significance lies in:

1. Quality Control: It serves as a rapid QC tool to verify that the correct strain is being analyzed and to detect unexpected contaminants before deep proteomic analysis.
2. Strain-Specific Analysis: By retrieving the closest matching strain from GTDB, it ensures downstream gene content analysis accurately reflects the specific organism, which is crucial for prokaryotes where strain variation is high.
3. Microbial Diversity Discovery: It enables the identification of dominant members in complex communities without the need for 16S rRNA or whole-genome sequencing, offering a protein-centric view of microbial composition.

In summary, NovoTax transforms raw proteomics data into actionable taxonomic insights, bridging the gap between mass spectrometry and microbial genomics.