# Taxonomic-Level Protein Quantification in Metaproteomics Using a Biomass-Constrained Expectation–Maximization Approach

## TL;DR

This paper introduces a new method to accurately quantify proteins from microbial communities using a modified algorithm that solves the shared peptide problem in metaproteomics.

## Contribution

The novel contribution is a biomass-constrained expectation–maximization algorithm integrated into the MiCId workflow to resolve taxon–protein quantification challenges.

## Key findings

- The algorithm accurately quantifies taxon–protein pairs in synthetic datasets with known species abundances.
- It effectively redistributes peptide counts among shared taxon–protein pairs in complex microbial datasets.
- Results from clinical stool datasets align with prior findings, confirming the method's accuracy in real-world microbiome analysis.

## Abstract

Microbiome communities are found across diverse environments
and
play critical roles in both ecosystem function and human health. Mass-spectrometry-based
metaproteomics provides a powerful means for directly identifying
and quantifying microbial proteins. However, its application is hindered
by the shared peptide problem, where peptides map to multiple proteins
across taxa, complicating taxon–protein quantification. To
address this challenge, we extend a previously published modified
expectation–maximization algorithm that incorporates taxonomic
biomass constraints into the Microorganism Classification and Identification
(MiCId) workflow. This enhanced expectation–maximization algorithm
is used to quantify taxon–protein pairs derived from clusters
of identified taxon–protein pairs, thereby enabling more accurate
quantification and representation of taxonomic-level proteomes. The
performance of the approach is evaluated using synthetic datasets
consisting of simple mixtures with known relative species abundances,
a more complex 24-species synthetic dataset, and a clinical human
stool microbiome dataset. It is shown that, in simple synthetic datasets,
fold changes computed for species–protein pairs closely match
the expected values and are consistent with those obtained from MaxQuant.
Using the 24-species synthetic dataset, we show that the algorithm
accurately redistributes peptide extracted ion count among taxon–protein
pairs that share peptides. Finally, analyzing the clinical stool microbiome
dataset, we demonstrate that MiCId’s results are accurate and
consistent with previously reported findings. These results demonstrate
the robustness of MiCId’s algorithm for quantifying taxon–protein
pairs in complex microbial communities. By resolving the shared peptide
problem, the method enables accurate representation of taxonomic-level
proteomes, thereby advancing the application of metaproteomics in
microbiome research.