MBG — Match-Between-Glycans

MBG is a lightweight, modular tool that expands glycopeptide identification in mass spectrometry-based glycoproteomics by inferring additional glycoforms at the MS1 level. It is fully integrated into FragPipe and designed to work seamlessly after an initial MSFragger-Glyco database search.

---

Overview

Protein glycosylation is one of the most important and complex post-translational modifications, regulating protein function, stability, and localization. Because glycosylation is a non-template-driven biosynthetic process, a single glycosite typically carries many co-occurring glycoforms that differ by the sequential addition or removal of monosaccharide units. This dilutes the signal of individual glycopeptides, leading low-abundance glycoforms to be missed in standard DDA database searches.

MBG addresses this by exploiting a key property of reverse-phase LC separation: glycopeptides sharing the same peptide backbone elute at nearly the same retention time (RT), regardless of their glycan. After an initial MSFragger-Glyco search provides a set of high-confidence glycopeptide identifications, MBG:

1. Generates candidate glycoforms — for each identified glycopeptide, it predicts neighboring glycoforms differing by one or more monosaccharide units (e.g., +Hex, +HexNAc, +NeuAc, +Fuc) or user-defined glycan modifications (e.g., NH₄⁺ or Fe³⁺ adducts).
2. Searches for MS1 evidence — using IonQuant, it looks for precursor signals at the expected m/z and within a narrow RT/IM window anchored to the parent glycopeptide's observed RT/IM plus a learned monosaccharide-specific shift.
3. Scores and filters candidates — a linear discriminant analysis (LDA) model with 7 features (RT/IM shift, mass error, precursor intensity, Y0/Y1 ion relative intensities, isotope envelope KL divergence, glycan shift frequency) separates targets from decoys (decoys use a +11 Da mass offset). FDR is controlled at the precursor level.

The result is an expanded PSM table containing both the original MSFragger-Glyco identifications and the new MBG-inferred glycopeptides, providing a more complete quantitative profile of glycosylation at each glycosite.

---

Key Features

• MS1-based inference — identifies glycopeptides that lack MS2 spectra or have low-quality MS2, using only precursor-level evidence.
• Learned RT/IM shifts — monosaccharide-specific RT and ion-mobility (IM) shifts are estimated from the data itself, making MBG robust across different LC gradients and instrument platforms.
• Adduct and modification recovery — can recover glycoforms bearing adducts (NH₄⁺, Fe³⁺, Na⁺) or modifications (e.g., phosphorylation for M6P glycans) without expanding the original database search space.
• Target-decoy FDR control — rigorous statistical filtering at a user-defined FDR threshold.
• FragPipe integration — runs as a one-click step inside the FragPipe glycoproteomics workflow, compatible with label-free and TMT-labeled experiments, DDA and PASEF data.

---

Workflow

Raw files
   │
   ▼
MSFragger-Glyco (database search)
   │
   ▼
PTM-Shepherd + Philosopher (PSM validation, FDR filtering)
   │
   ▼
IonQuant (quantification of identified glycopeptides)
   │
   ▼
MBG  ◄─── this tool
   │   1. Group PSMs by glycosite
   │   2. Estimate per-monosaccharide RT/IM shifts
   │   3. Generate candidate glycoforms (+/- monosaccharides, adducts)
   │   4. Search MS1 via IonQuant within RT/IM windows
   │   5. Score with LDA; apply FDR filter
   │
   ▼
Expanded PSM table (original + inferred glycopeptides)
   │
   ▼
Downstream analysis / FragPipe-Analyst

---

Performance (from the manuscript)

Dataset	Instrument	Increase in glycopeptide IDs
Fission yeast	Orbitrap Fusion	+7.5% at 1% FDR; +23.6% at 5% FDR
Human plasma (PASEF)	timsTOF HT	+14.6%
GBM (CPTAC, TMT-11)	Orbitrap Fusion Lumos	740 exclusive MBG IDs (4.6% of total)
Mouse liver (adducts)	Orbitrap Fusion	+1,234 glycopeptides incl. NH₄⁺ and Fe³⁺ adducts

Entrapment analysis on the yeast dataset showed an estimated false inference rate of 0.63%.

---

Requirements

• Java 11 or later
• FragPipe (for integrated use) — MBG is bundled as a tool within FragPipe
• IonQuant (bundled as a dependency) — used internally for MS1 feature detection
• BatMassIO (bundled) — used for reading raw spectrum files (Thermo .raw, Bruker .d, mzML, etc.)

Building from source

./gradlew jar

The output JAR will be at build/libs/MBG-<version>.jar.

Installation

Via FragPipe (recommended):

MBG is bundled with FragPipe. Simply download and install FragPipe — no separate installation of MBG is needed.

Standalone:

1. Install Java 11 or later.
2. Download the latest mbg-<version>.jar from the Releases page, or build from source with ./gradlew jar.

Typical install time: Less than 5 minutes (as part of FragPipe installation on a standard desktop computer). Building from source takes approximately 1–2 minutes.

---

Usage

MBG is typically run automatically within FragPipe. For standalone use:

java -jar mbg-<version>.jar --match [args]

Required arguments

Argument	Description
--psm <path>	Path to the input PSM file (FragPipe psm.tsv format)

Optional arguments

Argument	Default	Description
--manifest <path>	—	Path to the FragPipe .fp-manifest file mapping raw file names to full paths
--residuedb <path>	built-in	Path to a custom glycan residue definitions file
--glycanmoddb <path>	built-in	Path to a custom glycan modification definitions file (for adducts, etc.)
--maxq <float>	0.01	Maximum glycan q-value to accept as a high-confidence input glycoPSM
--minpsms <int>	2	Minimum number of PSMs required for a glycan to be used for RT/IM shift estimation
--minglycans <int>	2	Minimum number of distinct glycans observed per peptide to enable inference
--rttol <float>	0.4	RT tolerance (minutes) for matching inferred glycoforms
--imtol <float>	0.05	Ion mobility tolerance (V·s·cm⁻²) for PASEF data
--mztol <int>	5	Mass tolerance (ppm) for MS1 matching
--fdr <float>	0.05	FDR threshold for accepting inferred glycopeptides
--nopasef <bool>	false	Set to true for non-PASEF (Orbitrap) data
--runtmt <bool>	false	Set to true for TMT-labeled experiments
--toaddresiduals <str>	—	Comma-separated list of monosaccharides or glycan compositions to add as candidate shifts (e.g., HexNAc,Fuc,HexNAc(1)Hex(1))
--expanddb <int>	0	Number of additional rounds of iterative inference on newly found peaks
--maxskips <int>	0	Number of missed peaks allowed during iterative expansion
--numthreads <int>	4	Number of threads for parallel processing
--allowchimeric <bool>	false	Allow chimeric spectra when searching for supporting MS2

Example

java -jar mbg-0.3.7.jar --match \
    --psm /data/experiment/psm.tsv \
    --manifest /data/experiment/fragpipe.fp-manifest \
    --maxq 0.01 \
    --minpsms 1 \
    --rttol 0.4 \
    --fdr 0.05 \
    --nopasef true

---

Demo: Fission Yeast Dataset

This demo walks through running MBG on the fission yeast glycoproteomics dataset (PXD005565), a well-characterized benchmark used in the manuscript.

1. Download the demo data

Download the fission yeast dataset from ProteomeXchange:

• Accession: PXD005565
• Download the .raw files (3 files).
• (optional) Convert to mzML using MSConvert.

2. Run FragPipe with MSFragger-Glyco and MBG

1. Open FragPipe and load the mzML files.

2. Select the glyco-N-HCD workflow.

3. Click Database panel to set the protein database to the S. pombe proteome (download from UniProt, Proteome ID UP000002485, add decoys and contaminants).

4. Click Glyco panel set the glycan database as Mouse_N-glycans-1670-pGlyco. Check Run Glycoform Inference and adjust MBG setting through MS1 Glycoform Inference block as the following:

   • Max Glycan q-value: 0.01
   • Min PSMs: 1
   • Min Glycans: 2
   • MBG FDR: 0.01
   • Max Inference Step: 2
   • Max Inference Skips: 0
   • Allow inferring chimeric spectra: checked
   • Residuals to add: Hex(1),NH4(1)

5. Click Quant(MS1) panel, in Feature dectection and peak tracing block, set m/z tolerance to 10, RT tolerance to 0.4

6. Run Fragpipe

3. Run MBG out of Fragpipe

Run Fragpipe as the above parameters without checking Run Glycoform Inference, after the FragPipe search completes, run MBG on the output:

java -jar mbg-0.3.7.jar --match \
    --psm /path/to/yeast_output/psm.tsv \
    --manifest /path/to/yeast_output/fragpipe.fp-manifest \
    --toaddresiduals Hex(1),NH4(1) \
    --maxq 0.01 \
    --minpsms 1 \
    --minglycans 2 \
    --rttol 0.4 \
    --mztol 10 \
    --fdr 0.01 \
    --expanddb 2 \
    --nopasef true

4. Expected output

After MBG completes, you should see:

• Updated psm.tsv — the original PSMs plus MBG-inferred glycopeptides. At 5% FDR, expect approximately a 7.5% (or 23.6% with 5% MBG FDR) increase in unique glycopeptide identifications.
• rt_shifts.csv — retention time shifts per glycan transition (e.g., HexNAc(2)Hex(8) → HexNAc(2)Hex(9)). Most high-mannose transitions should fall within ~0.4 minutes.
• MBG-inferred entries are flagged in the PSM table and can be distinguished from MSFragger-Glyco identifications for downstream analysis.

5. Expected run time

• FragPipe search (MSFragger-Glyco + PTM-Shepherd + IonQuant): approximately 15–20 minutes for 3 raw files on a standard desktop.
• MBG step only: approximately 2–5 minutes on the same hardware.

---

Output

MBG appends inferred glycopeptide PSMs to the input PSM table and writes:

• Updated psm.tsv — original PSMs plus MBG-inferred entries (marked with an MBG flag in the Hyperscore or source columns)
• rt_shifts.csv / im_shifts.csv — per-monosaccharide RT and IM shift distributions (useful for QC and plotting)
• Skyline modifications file — definitions for any novel glycan compositions inferred by MBG, for import into Skyline

---

Citation

Shen J, Polasky DA, Jager S, Yu F, Heck AJR, Reiding KR, Nesvizhskii AI. Expanding Glycopeptide Identification with Match-Between-Glycans in FragPipe. Manuscript in preparation, 2026. Data: PRIDE PXD074575.

---

License

See LICENSE for details.