mosbi

Workflows

Standard Workflow

Run multiple biclustering algorithms on a data matrix, construct a similarity network with an error model, extract robust communities, and generate consensus ensemble biclusters.

library(mosbi)

# 1. Prepare data matrix
data(mouse_data)
mouse_data_sub <- mouse_data[c(grep("metabolite_identification", colnames(mouse_data)), grep("^X", colnames(mouse_data)))]
# Simple mock matrix for runnable demonstration
data_matrix <- matrix(rnorm(1000), nrow = 100, ncol = 10)
rownames(data_matrix) <- paste0("M", 1:100)
colnames(data_matrix) <- paste0("S", 1:10)

# 2. Run Biclustering Algorithms
fb <- mosbi::run_fabia(data_matrix)
BCisa <- mosbi::run_isa(data_matrix)
BCplaid <- mosbi::run_plaid(data_matrix)
BCqubic <- mosbi::run_qubic(data_matrix)
all_bics <- c(fb, BCisa, BCplaid, BCqubic)

# 3. Compute Bicluster Similarity Network with Error Model
bic_net <- mosbi::bicluster_network(all_bics, data_matrix, n_randomizations = 5, MARGIN = "both", metric = 4)

# 4. Extract Louvain Communities
coms <- mosbi::get_louvain_communities(bic_net, min_size = 3, bics = all_bics)

# 5. Generate Consensus Ensemble Biclusters
ensemble_bicluster_list <- mosbi::ensemble_biclusters(coms, all_bics, data_matrix, row_threshold = 0.1, col_threshold = 0.1)

Input: A numeric data matrix; Output: A list of robust ensemble biclusters.

When to Use

• When you want to run multiple biclustering algorithms (e.g., Fabia via run_fabia, ISA via run_isa, Plaid via run_plaid, QUBIC via run_qubic) and combine their results.
• When you need to compute a similarity network of biclusters using metrics like Fowlkes-Mallows index, Jaccard index, Bray-Curtis similarity, or overlap coefficient via bicluster_network.
• When you want to extract robust bicluster communities using Louvain modularity via get_louvain_communities and generate consensus ensemble biclusters via ensemble_biclusters.

When NOT to Use

• For standard single-algorithm biclustering without ensemble combination, where direct packages like biclust or fabia can be used directly.
• When you do not have a numeric data matrix (e.g., categorical or purely qualitative data).

Data Requirements

• A numeric data matrix (e.g., data_matrix in the vignette derived from mouse_data using log2 and z-score normalization).
• No missing values are preferred (as noted: "The data has a gaussian-like distribution and no missing values").

Key Parameters

• n_randomizations (5): Number of randomizations for the error model in bicluster_network.
• MARGIN ("both"): Use datapoints for metric evaluation in bicluster_network.
• metric (4): Similarity metric index (e.g., 4 for Fowlkes-Mallows index) in bicluster_network.
• n_steps (1000): Number of steps at which the cut-off is evaluated in bicluster_network.
• plot_edge_dist (TRUE): Plot the evaluation of cut-off estimation in bicluster_network.
• min_size (3): Minimum size of communities (number of biclusters) to save in get_louvain_communities.
• row_threshold (0.1): Minimum occurrence of a row-element in the biclusters of a community to be included in the ensemble bicluster.
• col_threshold (0.1): Minimum occurrence of a column-element in the biclusters of a community to be included in the ensemble bicluster.

Best Practices

• Normalize the input data matrix (e.g., using log2 transformation and z-score scaling) before running biclustering.
• Visualize the size distribution of the generated biclusters using colhistogram and rowhistogram (or a helper like bicluster_histo).
• Evaluate the cut-off estimation for the similarity network by setting plot_edge_dist = TRUE in bicluster_network.
• Visualize the bicluster similarity network using plot or plot_algo_network to inspect how different algorithms contribute to communities.

Common Pitfalls

• Running get_louvain_communities with a min_size that is too high, resulting in zero saved communities. Fix: Lower min_size (e.g., to 3).
• Biclustering algorithms throwing errors or returning empty lists. Fix: Ensure the input data matrix has a gaussian-like distribution and no missing values.

Alternatives

• biclust for running individual biclustering algorithms like Plaid.
• fabia for Fabia biclustering.
• QUBIC for QUBIC biclustering.
• isa2 for ISA biclustering.

Citations

• Rose TD (2025). MoSBi: Molecular signature Identification from Biclustering.

References

• Homepage: bioconductor.org/packages/mosbi
• Vignette: https://bioconductor.org/packages/release/bioc/vignettes/mosbi/inst/doc/example-workflow.html