seroCOP

seroCOP is an R package for analysing correlates of protection using Bayesian methods. It fits generalized four-parameter logistic functions to antibody titre and infection outcome data using Stan, providing comprehensive model diagnostics and performance metrics.

Installation

You can install the development version of seroCOP from GitHub:

# Install devtools if needed
install.packages("devtools")

# Install seroCOP
devtools::install_github("seroanalytics/seroCOP")

Note: This package requires a working installation of brms and Stan. The brms package will handle Stan installation automatically when you first use it.

Quick Start

library(seroCOP)

# Simulate data
sim_data <- simulate_serocop_data(
  n = 200,
  floor = 0.05,
  ceiling = 0.90,
  ec50 = 1.5,
  slope = 2.0,
  seed = 123
)

# Create SeroCOP object
model <- SeroCOP$new(
  titre = sim_data$titre,
  infected = sim_data$infected
)

# Fit the model (uses brms with sensible defaults)
model$fit_model(chains = 4, iter = 2000)

# Get performance metrics
metrics <- model$get_metrics()

# Visualize results
model$plot_curve()
model$plot_roc()

# Extract parameter estimates
model$summary()

Hierarchical Group Effects

Model group-level heterogeneity (e.g., age groups with different correlates):

# Simulate age-stratified data
age_group <- sample(c("Young", "Middle", "Old"), 200, replace = TRUE)

# Fit hierarchical model
hier_model <- SeroCOP$new(
  titre = sim_data$titre,
  infected = sim_data$infected,
  group = age_group  # Adds random effects on slope and ec50
)

hier_model$fit_model(chains = 4, iter = 2000)

# Extract group-specific parameters
group_params <- hier_model$extract_group_parameters()

# Plot group-specific curves
hier_model$plot_group_curves()

Multi-Biomarker Analysis

Compare multiple biomarkers simultaneously with optional hierarchical effects:

# Prepare multi-biomarker data (matrix with columns = biomarkers)
titre_matrix <- cbind(
  IgG = rnorm(200, 2, 1),
  IgA = rnorm(200, 1.5, 1),
  Neutralization = rnorm(200, 3, 1.2)
)

# Initialize and fit (without groups)
multi_model <- SeroCOPMulti$new(
  titre = titre_matrix,
  infected = infected
)
multi_model$fit_all(chains = 4, iter = 2000)

# Compare biomarkers
multi_model$compare_biomarkers()
multi_model$plot_comparison()  # AUC vs LOO-ELPD plot
multi_model$plot_all_curves()

# With hierarchical effects across biomarkers
hier_multi <- SeroCOPMulti$new(
  titre = titre_matrix,
  infected = infected,
  group = age_group
)
hier_multi$fit_all(chains = 4, iter = 2000)
hier_multi$plot_group_curves()  # Shows all biomarkers × groups
hier_multi$extract_group_parameters()

Model

The package fits a four-parameter logistic model using brms (Bayesian Regression Models using Stan):

$$P(\text{infection} | \text{titre}) = \text{ceiling} \times \left[\frac{1}{1 + e^{\text{slope} \times (\text{titre} - \text{ec50})}} \times (1 - \text{floor}) + \text{floor}\right]$$

Where:

• floor: Lower asymptote (baseline infection probability at high titres)
• ceiling: Upper asymptote (maximum infection probability at low titres)
• ec50: Titre at 50% between floor and ceiling (inflection point)
• slope: Steepness of the dose-response curve

Hierarchical Extension

When group variables are provided, the model adds random intercepts:

$$\text{ec50}_g \sim \mathcal{N}(\text{ec50}_{\text{population}}, \sigma_{\text{ec50}})$$ $$\text{slope}_g \sim \mathcal{N}(\text{slope}_{\text{population}}, \sigma_{\text{slope}})$$

This allows for group-specific parameters while borrowing strength across groups.

Priors

The package automatically sets weakly informative priors based on your data:

• floor ~ Beta(1, 9): Favors low baseline infection (mean ≈ 0.1)
• ceiling ~ Beta(9, 1): Favors high maximum infection (mean ≈ 0.9)
• ec50 ~ Normal(μ, σ): Centered at midpoint of titre range
• slope ~ Normal(0, 2): Weakly informative, truncated at 0
• Group-level SDs ~ student_t(3, 0, 2.5): Weakly informative for hierarchical models

Documentation

See the package vignettes for detailed examples:

• SeroCOP Overview: Basic usage and model details
• Simulation and Recovery: Parameter recovery validation
• Multi-Biomarker Analysis: Comparing multiple antibody types
• Hierarchical Group Effects: Age-stratified and group-specific modeling

# View available vignettes
browseVignettes("seroCOP")

Performance Metrics

The package calculates several performance metrics:

• ROC AUC: Area under the ROC curve (discrimination)
• Brier Score: Mean squared error of predictions (calibration)
• LOO-CV: Leave-one-out cross-validation (out-of-sample performance)
• ELPD: Expected log pointwise predictive density

Citation

If you use this package in your research, please cite:

Hodgson, D. (2025). seroCOP: Correlates of Protection Analysis Using Bayesian Methods.
R package version 0.1.0.

Development

This package is under active development. Contributions and feedback are welcome.

Support

For questions, issues, or feature requests, please open an issue on GitHub.