PBLMM

Peptide-Based Linear Mixed Model for Proteomics Data Analysis

Version 2.1.1 (2023-10-23)

PBLMM is a Python package that implements a peptide-based linear mixed model approach for analyzing proteomics data. It allows for robust statistical analysis of protein abundance changes across different conditions by properly accounting for peptide-level variation.

Features

• Peptide-based statistical modeling using linear mixed models (LMM)
• Multiple protein rollup methods (sum, median, mean)
• Hypothesis testing with both LMM and t-test approaches
• Multiple testing correction using FDR control
• Flexible data processing compatible with various proteomics data formats

Installation

pip install git+https://github.com/science64/PBLMM.git

Quick Start

import pandas as pd
from PBLMM import HypothesisTesting, Defaults

# Initialize defaults
defaults = Defaults()

# Load your proteomics data (peptide or PSM level)
data = pd.read_csv("your_proteomics_data.csv")

# Define your experimental conditions
conditions = ["Control", "Treatment", "Control", "Treatment"]

# Define pairs for comparison
pairs = [["Treatment", "Control"]]

# Initialize hypothesis testing
ht = HypothesisTesting(defaults)

# Run peptide-based LMM analysis
results = ht.peptide_based_lmm(
    input_file=data,
    conditions=conditions,
    pairs=pairs
)

# Save results
results.to_csv("pblmm_results.csv")

Understanding the Statistical Output

P-values

The PBLMM code generates p-values using two different statistical approaches:

1. Linear Mixed Models (LMM)

In the peptide_based_lmm method:

• P-values are generated by fitting a linear mixed model to peptide-level data
• The model uses peptide sequences as random effects (groups='Sequence')
• The model formula used is: "value ~ variable" where:
  • value represents log2-transformed abundance measurements
  • variable represents the experimental conditions being compared
• P-values test the significance of the condition effect (null hypothesis: no difference between conditions)

2. Unpaired t-test

In the ttest method:

• P-values come from a standard two-sample t-test with equal variance assumption
• For each protein, the test compares abundance values between two specified conditions
• Tests whether the means of the two conditions are significantly different

Q-values

Q-values are only calculated in the peptide_based_lmm method:

• They represent p-values adjusted for multiple hypothesis testing
• Correction is performed using the Benjamini-Hochberg procedure (FDR correction)
• Q-values control the false discovery rate (FDR), which is the expected proportion of false positives among all significant results
• This adjustment is critical when testing many proteins simultaneously to minimize false discoveries

Advanced Usage

Custom Column Names

You can customize the column names used for protein accessions, sequences, and abundance values:

defaults = Defaults()
defaults.MasterProteinAccession = "Your Protein ID Column"
defaults.sequence = "Your Sequence Column"
defaults.AbundanceColumn = "Your Abundance Column Prefix"

Technical Replicates and Multiplexing

For experiments with technical replicates or multiple plexes:

results = ht.peptide_based_lmm(
    input_file=data,
    conditions=conditions,
    techreps=["rep1", "rep1", "rep2", "rep2"],
    plexes=["plex1", "plex1", "plex2", "plex2"],
    pairs=pairs
)

Citation

If you use PBLMM in your research, please cite:

Original implementation by Kevin Klann
Updated by Süleyman Bozkurt

License

MIT

Contributors

• Kevin Klann (Original author)
• Süleyman Bozkurt (Maintainer)