MOFSR

Documentation: https://zaoqu-liu.github.io/MOFSR/

Multi-Omics Fusion for Subtype Recognition

MOFSR is a comprehensive R package designed for integrative analysis of multi-omics data to identify disease subtypes and characterize molecular heterogeneity. The package implements 15 state-of-the-art multi-omics clustering algorithms with internal implementations to ensure maximum cross-platform compatibility and reproducibility.

Key Features

15 Multi-Omics Integration Algorithms: SNF, wSNF, CPCA, iClusterBayes, IntNMF, LRAcluster, MCIA, MOFA, NEMO, PINSPlus, RGCCA, SGCCA, CIMLR, BCC, and Late Fusion
17 Classification Methods: Including SVM, Random Forest, XGBoost, Neural Networks, and ensemble approaches
Parallel Computing: Built-in support via the future framework for large-scale analyses
Comprehensive Visualization: UMAP, consensus heatmaps, silhouette plots, and survival curves
Data Preprocessing: Normalization, feature filtering, missing value imputation, and batch correction
Cluster Quality Assessment: PAC, CHI, silhouette width, and ARI metrics

Methodological Overview

MOFSR provides a unified interface for multi-omics subtyping through three major analytical paradigms:

Paradigm	Algorithms	Description
Early Fusion	CPCA, MCIA, RGCCA, SGCCA	Concatenate or jointly decompose data matrices before clustering
Late Fusion	BCC, Late Fusion, PINSPlus	Cluster each omics separately, then combine results
Intermediate Fusion	SNF, wSNF, CIMLR, NEMO	Construct similarity networks and fuse at the network level
Model-Based	MOFA, iClusterBayes, IntNMF, LRAcluster	Probabilistic or matrix factorization approaches

Installation

From R-Universe (Recommended)

install.packages("MOFSR", repos = "https://zaoqu-liu.r-universe.dev")

From GitHub

# install.packages("remotes")
remotes::install_github("Zaoqu-Liu/MOFSR")

Quick Start

library(MOFSR)

# Prepare multi-omics data (features × samples)
data_list <- list(
  mRNA = mRNA_matrix,
  miRNA = miRNA_matrix,
  methylation = methylation_matrix
)

# Run SNF clustering
result <- run_snf(data_list, n_clusters = 3)

# Or use the unified interface
result <- run_integration(data_list, algorithm = "SNF", n_clusters = 3)

# Compare multiple algorithms
algorithms <- c("SNF", "RGCCA", "CIMLR", "MOFA")
results <- lapply(algorithms, function(alg) {
  run_integration(data_list, algorithm = alg, n_clusters = 3)
})
names(results) <- algorithms

# Evaluate clustering consistency
ari_matrix <- compare_clusterings(results)

Whole-slide pathology data

MOFSR does not read .svs, .ndpi, .scn, or .mrxs pixels directly. Extract slide-level numeric measurements or embeddings first, then supply a features × samples matrix as the pathology modality. The WSI to MOFSR workflow gives a reproducible QuPath export, aggregation, quality-control, and sample-alignment recipe.

Supported Algorithms

Algorithm	Reference	Year
SNF	Wang et al., Nature Methods	2014
CIMLR	Ramazzotti et al., Nature Communications	2018
iClusterBayes	Mo et al., Biostatistics	2018
IntNMF	Chalise & Fridley, PLoS ONE	2017
RGCCA/SGCCA	Tenenhaus & Tenenhaus, Psychometrika	2011
NEMO	Rappoport & Shamir, Bioinformatics	2019
PINSPlus	Nguyen et al., Genome Research	2017
MCIA	Meng et al., BMC Bioinformatics	2014
LRAcluster	Wu et al., Cancer Informatics	2015
BCC	Lock & Dunson, Bioinformatics	2013

Data Preprocessing

# Normalize data
data_norm <- normalize_omics(data_list, method = "zscore")

# Filter low-variance features
data_filtered <- filter_low_variance(data_list, min_var = 0.01)

# Handle missing values
data_imputed <- handle_missing(data_list, method = "knn", k = 5)

# Batch correction
data_corrected <- correct_batch(data_list, batch = batch_labels)

# Quality control summary
qc_report <- qc_summary(data_list)

Visualization

# UMAP visualization
umap_coords <- compute_umap(data_list)
plot_umap(umap_coords, result$Cluster)

# Consensus matrix heatmap
cc_result <- RunCC(data_list[[1]], maxK = 6)
plot_consensus_heatmap(cc_result[[3]]$consensusMatrix)

# Silhouette analysis
plot_silhouette(result$Cluster, dist(t(data_list[[1]])))

# Survival analysis (requires survival data)
plot_survival(time, event, result$Cluster)

Parallel Computing

library(future)
plan(multisession, workers = 4)

# Parallel consensus clustering
result <- parallel_consensus_cluster(data, max_k = 6, n_reps = 1000)

# Parallel feature selection
features <- parallel_bootstrap_features(data, n_bootstrap = 100)

System Requirements

R ≥ 4.0.0
Operating System: Windows, macOS, Linux
Core Dependencies: stats, graphics, utils (base R), and dplyr
Optional: ggplot2, survival, future (for enhanced functionality)

All 15 clustering algorithms are implemented internally without external dependencies, ensuring reproducibility across different computing environments.

Citation

If you use MOFSR in your research, please cite:

Liu Z (2024). MOFSR: Multi-Omics Fusion for Subtype Recognition. 
R package version 2.3.0, https://github.com/Zaoqu-Liu/MOFSR

Related Publications

The algorithms implemented in MOFSR are based on peer-reviewed publications. Please also cite the original algorithm papers when using specific methods.

Author

Zaoqu Liu

License

GPL-3.0