BiSSLB

BiSSLB: Binary Spike-and-Slab Lasso Biclustering for binary matrices

BiSSLB is an R package for performing sparse biclustering on binary matrices. It uses spike-and-slab lasso priors on matrix factorizations to reveal interpretable biclusters. The core computations are accelerated using Rcpp and RcppArmadillo.

---

🔧 Installation

To install the development version of the package from GitHub:

# Install devtools if not already installed
install.packages("devtools")

# Install BiSSLB from GitHub
devtools::install_github("sijianf/BiSSLB")

---

📦 Package Overview

The BiSSLB package includes the following key functions:

Function	Description
BiSSLB()	Main interface to run biclustering on a binary matrix
BiSSLB_ladder()	Run BiSSLB across a grid of spike prior parameters
main_iterations()	Core C++ engine performing iterative updates (not user-facing)
get_thetas()	Estimate sparsity parameters under spike-and-slab prior
get_logLikelihood()	Compute the log-likelihood and BIC of current model state

---

🚀 Quick Start

library(BiSSLB)

# Simulate a binary matrix
set.seed(123)
Y <- matrix(rbinom(100, 1, 0.4), nrow = 10)

# Run BiSSLB with default settings
result <- BiSSLB(Y, K_init = 5)

# Inspect output
names(result)

# View estimated row/column profiles
image(result$A)
image(result$B)

---

🔁 Parameter Tuning Example

Use BiSSLB_ladder() to explore multiple regularization settings:

out_ladder <- BiSSLB_ladder(
  Y,
  tilde_lambda_0s = c(5, 10),
  lambda_0s = c(5, 10),
  tilde_lambda_1 = 1,
  lambda_1 = 1,
  K_init = 10
)

# Access results and BICs
str(out_ladder$results)
out_ladder$BICs

---

👉 Method Summary

BiSSLB models a binary matrix Y as:

$$p(y_{ij} = 1) = \text{sigmoid}(a_i^T b_j + \mu_i)$$

The matrices A and B are constrained by spike-and-slab lasso priors to induce sparsity. An optional IBP-style reordering helps emphasize interpretable structures during iteration.

---

📘 Documentation

You can view the documentation for all exported functions via:

?BiSSLB
?BiSSLB_ladder

---

📄 License

This package is licensed under the GPL-3 License. See the LICENSE file for details.