SMAD: Statistical Modelling of AP-MS Data

SMAD is an R package designed for the statistical analysis of Affinity Purification–Mass Spectrometry (AP-MS) data. It implements several widely-used algorithms to compute confidence scores, helping researchers distinguish bona fide protein-protein interactions (PPI) from non-specific background noise.

Table of Contents

• Installation
• Input Data Format
• Available Scoring Algorithms
• Quick Start
• References

Installation

You can install the development version of SMAD from GitHub using devtools or remotes:

# Install devtools if not already installed
if (!requireNamespace("devtools", quietly = TRUE)) install.packages("devtools")

# Install SMAD
devtools::install_github("zqzneptune/SMAD")

# Load the package
library(SMAD)

Input Data Format

To ensure compatibility with SMAD scoring functions, your input data.frame should follow a specific structure. You can explore the built-in demo dataset:

data(TestDataInput)
head(TestDataInput)

Required Columns

Column	Description
idRun	Unique identifier for a specific AP-MS run (replicate).
idBait	Unique identifier for the bait protein.
idPrey	Unique identifier for the prey protein.
countPrey	Spectral counts (or peptide counts) for the prey protein.
lenPrey	Protein sequence length of the prey (required for HGScore/NSAF).

Note on Replicates: If you have biological or technical duplicates, ensure the idRun is unique (e.g., append suffixes like _A, _B). The combination of idRun and idBait should uniquely identify a single purification.

---

Available Scoring Algorithms

1. CompPASS

The Comparative Proteomic Analysis Software Suite (CompPASS) uses a "spoke model" to identify high-confidence interactions. It outputs several metrics, including Z-score, S-score, D-score, and WD-score. This implementation is optimized for performance based on the BioPlex pipeline.

• Required columns: idRun, idBait, idPrey, countPrey.
• Usage: datScore <- CompPASS(datInput)

2. DICE

The Dice coefficient measures the similarity between prey pair-wise combinations. It is particularly useful for identifying preys that frequently co-occur across different runs.

• Required columns: idRun, idPrey.
• Usage: datScore <- DICE(datInput)

3. Hart

Based on a hypergeometric distribution error model, this algorithm calculates the probability of finding a prey across different bait purifications by chance.

• Required columns: idRun, idPrey.
• Usage: datScore <- Hart(datInput)

4. HGScore

HGScore enhances the Hart hypergeometric model by incorporating NSAF (Normalized Spectral Abundance Factor), which accounts for protein length and abundance. It is designed based on a "matrix model."

• Required columns: idRun, idPrey, countPrey, lenPrey.
• Usage: datScore <- HG(datInput)

5. PE (Purification Enrichment)

The PE score incorporates both spoke and matrix models to calculate an interaction score based on the frequency and exclusivity of prey identifications.

• Required columns: idRun, idBait, idPrey.
• Usage: datScore <- PE(datInput)

---

Quick Start

library(SMAD)

# Load example data
data(TestDataInput)

# Run CompPASS scoring
results <- CompPASS(TestDataInput)

# View top-scoring interactions based on WD-score
head(results[order(results$WD, decreasing = TRUE), ])

License

Released under the MIT License.
© Qingzhou Zhang

References

1. CompPASS: Sowa et al., 2009 (Cell)
2. BioPlex: Huttlin et al., 2015 (Cell)
3. Hart Scoring: Hart et al., 2007 (BMC Bioinformatics)
4. HGScore: Guruharsha et al., 2011 (Cell)
5. DICE: Zhang et al., 2008 (Bioinformatics)
6. PE Score: Collins et al., 2007 (MCP)