bioconductor-beer

name

bioconductor-beer

description

BEER implements a Bayesian model for analyzing phage-immunoprecipitation sequencing (PhIP-seq) data. Given a PhIPData object, BEER returns posterior probabilities of enriched antibody responses, point estimates for the relative fold-change in comparison to negative control samples, and more. Additionally, BEER provides a convenient implementation for using edgeR to identify enriched antibody responses.

beer

Workflows

Standard Workflow

Identify enriched antibody responses to peptides in PhIP-seq data using edgeR and Bayesian (BEER) modeling.

library(beer)

# Load simulated PhIPData
data_path <- system.file("extdata/sim_data.rds", package = "beer")
sim_data <- readRDS(data_path)

# 1. Run edgeR to estimate peptide-specific dispersion
edgeR_out <- runEdgeR(sim_data, assay.names = c(logfc = "edgeR_logfc", prob = "edgeR_logpval"))

# 2. Run BEER (brew) to identify enriched peptides (hits)
assay_locations <- c(phi = "beer_fc_marg", phi_Z = "beer_fc_cond", Z = "beer_prob", c = "sampleInfo", pi = "sampleInfo")
beer_out <- brew(edgeR_out, assay.names = assay_locations)

# Identify hits by combining posterior probabilities and super-enriched peptides
was_run <- matrix(rep(beer_out$group != "beads", each = nrow(beer_out)), nrow = nrow(beer_out))
are_se <- was_run & is.na(assay(beer_out, "beer_prob"))
assay(beer_out, "beer_hits") <- assay(beer_out, "beer_prob") > 0.5 | are_se

# 3. Run beads-only round robin to estimate false positive rates
beer_beadsRR <- beadsRR(beer_out, method = "beer", assay.names = assay_locations)

# 4. Calculate expected reads/proportions and Bayes factors
expected_data <- getExpected(sim_data)
bf_data <- getBF(beer_out)

Input: A PhIPData object containing PhIP-seq read counts; Output: A PhIPData object updated with enrichment hits, posterior probabilities, and Bayes factors.

When to Use

• Analyzing PhIP-seq (phage immunoprecipitation sequencing) data to identify enriched antibody responses to peptides.
• When you want to run standard differential expression analysis using runEdgeR on PhIP-seq data.
• When you want to apply a Bayesian hierarchical model via brew to estimate posterior probabilities of enrichment and relative fold-changes.
• When you want to estimate false positive rates using a beads-only round robin via beadsRR.

When NOT to Use

• For general RNA-seq differential expression without a PhIP-seq design or negative control (beads-only) samples. Use standard edgeR or DESeq2 directly.

Data Requirements

• A PhIPData object (e.g., loaded from sim_data.rds in the vignette) containing read counts for samples and peptides, with designated beads-only control samples.

Key Parameters

• assay.names (NULL): Named vector specifying where to store summarized MCMC output (e.g., phi, phi_Z, Z, c, pi).
• beadsRR (FALSE): Logical indicating whether to run beads-only round robin within brew or runEdgeR.
• de.method ("exactTest"): Method for edgeR differential expression, either "exactTest" or "glmQLFTest".
• a_pi (2): Prior shape parameter for the beta distribution describing the proportion of enriched peptides.
• b_pi (300): Prior shape parameter for the beta distribution describing the proportion of enriched peptides.
• a_phi (1.25): Prior shape parameter for the gamma distribution describing fold-change for enriched peptides.
• b_phi (0.1): Prior shape parameter for the gamma distribution describing fold-change for enriched peptides.
• fc (1): Minimum fold-change for an enriched peptide.

Best Practices

• Run runEdgeR first to quickly identify enriched peptides and provide a baseline comparison.
• Identify and exclude "super-enriched" peptides (e.g., fold-change > 15) before running the MCMC to speed up BEER execution.
• Use parallelization via BiocParallel::SerialParam() or BiocParallel::SnowParam() passed to BPPARAM to speed up the Bayesian model.
• Verify MCMC convergence by saving samples to a directory and plotting trace/density plots using coda or plot.

Common Pitfalls

• MCMC sampler getting stuck during execution. Fix: Ensure prior beta shape parameters (e.g., estimated via getAB) are greater than one.
• Extremely slow execution times for large datasets. Fix: Set a conservative threshold for removing super-enriched peptides to reduce the number of peptides modeled by MCMC.

Alternatives

• edgeR directly for standard negative binomial differential expression.
• DESeq2 for alternative parametric differential expression analysis.

Citations

• Chen A, Kammers K, Larman HB, Scharpf R, Ruczinski I (2022). Detecting antibody reactivities in phage immunoprecipitation sequencing data. bioRxiv.
• Robinson MD, McCarthy DJ and Smyth GK (2010). edgeR: a Bioconductor package for differential expression analysis of digital gene expression data. Bioinformatics.

References

• Homepage: bioconductor.org/packages/beer
• Vignette: https://bioconductor.org/packages/release/bioc/vignettes/beer/inst/doc/beer.html