B-cell-based Hybrid Immune Virtual Evolution (bHIVE) for caret

A wrapper for integrating the B-cell Hybrid Immune Variant Engine (bHIVE) algorithm with the caret package. Supports both classification and regression tasks, providing compatibility with caret::train() for model training and validation.

Usage

bHIVEmodel

Format

A list containing the components required for integration with the caret package.

Details

The bHIVEmodel wrapper facilitates the use of bHIVE for classification and regression. It defines the model label, parameter grid, fitting function, and prediction methods to conform to the caret model specification.

Components

label

Character string. Identifies the model as "B-cell-based Hybrid Immune Virtual Evolution".

library

Character string. Specifies the R package containing the bHIVE implementation. Default is "customPackage".

type

Character vector. Specifies the supported tasks: "Classification" and "Regression".

parameters

A data.frame describing the tunable parameters:

parameter: Name of the parameter.
class: Data type of the parameter ("numeric").
label: Short description of the parameter.

grid

Function. Generates a grid of tuning parameters for hyperparameter optimization.

fit

Function. Trains the bHIVE model using specified hyperparameters and task type.

predict

Function. Generates predictions for new data (classification labels or regression values).

prob

Function. Calculates class probabilities for classification tasks.

Parameters

nAntibodies: Number of initial antibodies in the bHIVE algorithm.
beta: Clone multiplier. Controls the number of clones generated for top-matching antibodies.
epsilon: Similarity threshold for antibody suppression. Smaller values encourage more diversity in the repertoire.

Functions

grid(x, y, len): Generates a grid of tuning parameters. Accepts:
- x: Feature matrix or data frame.
- y: Target vector (factor for classification, numeric for regression).
- len: Number of grid points for each parameter.
fit(x, y, wts, param, lev, last, classProbs, ...): Trains the bHIVE model. Key arguments:
- x: Feature matrix or data frame.
- y: Target vector.
- param: List of hyperparameters (nAntibodies, beta, epsilon).
- ...: Additional arguments passed to the bHIVE function.
predict(modelFit, newdata, submodels): Generates predictions for new data.
- modelFit: Trained bHIVE model.
- newdata: New feature data for prediction.
prob(modelFit, newdata, submodels): Calculates class probabilities (classification only).

Example Usage


library(caret)

# Simulated classification dataset
set.seed(123)
X <- matrix(rnorm(100 * 5), ncol = 5)
y <- factor(sample(c("Class1", "Class2"), 100, replace = TRUE))

# Train bHIVE model using caret
trainControl <- trainControl(method = "cv", number = 5, classProbs = TRUE)
tunedModel <- train(
  x = X,
  y = y,
  method = bHIVEmodel,
  trControl = trainControl,
  tuneLength = 3
)

# Predictions
predictions <- predict(tunedModel, newdata = X)
probabilities <- predict(tunedModel, newdata = X, type = "prob")

Examples

# View model structure
bHIVEmodel$label
#> [1] "Artificial Immune Network (bHIVE)"
bHIVEmodel$parameters
#>     parameter   class                label
#> 1 nAntibodies numeric Number of Antibodies
#> 2        beta numeric     Clone Multiplier
#> 3     epsilon numeric    Epsilon Threshold