bio-multi-omics-mofa-integration

name	bio-multi-omics-mofa-integration
description	Integrate multiple omics datasets with MOFA2 to discover shared and modality-specific latent factors driving biological variation. Use when: user wants to combine metabolomics with transcriptomics/proteomics, find multi-omics factors, identify shared variation across data types, or run unsupervised multi-omics integration. Triggers: multi-omics factors, shared variation, MOFA, MOFA2, multi-omics integration, latent factors, data modality integration, metabolomics + transcriptomics, factor analysis across omics.
tool_type	r
primary_tool	MOFA2
upstream	{"repo":"https://github.com/GPTomics/bioSkills","license":"MIT","original_author":"GPTomics"}

Version Compatibility

Reference examples tested with: scanpy 1.10+

Before using code patterns, verify installed versions match. If versions differ:

• R: packageVersion('<pkg>') then ?function_name to verify parameters

If code throws ImportError, AttributeError, or TypeError, introspect the installed package and adapt the example to match the actual API rather than retrying.

MOFA2 Integration

"Find shared variation across my omics layers" → Discover latent factors that capture shared and modality-specific sources of biological variation in an unsupervised manner.

• R: MOFA2::create_mofa() → prepare_mofa() → run_mofa()
• Python: mofapy2 for training, muon for downstream

Prepare Multi-Omics Data

Goal: Load and align multiple omics matrices into a consistent format for MOFA2 input.

Approach: Read each omics layer, intersect to common samples, transpose to features-by-samples orientation.

library(MOFA2)
library(MultiAssayExperiment)

# Load individual omics matrices (samples x features)
rna <- as.matrix(read.csv('rnaseq_matrix.csv', row.names = 1))
protein <- as.matrix(read.csv('proteomics_matrix.csv', row.names = 1))
methylation <- as.matrix(read.csv('methylation_matrix.csv', row.names = 1))

# Ensure consistent sample names across views
common_samples <- Reduce(intersect, list(rownames(rna), rownames(protein), rownames(methylation)))
rna <- rna[common_samples, ]
protein <- protein[common_samples, ]
methylation <- methylation[common_samples, ]

# Transpose to features x samples (MOFA format)
data_list <- list(
    RNA = t(rna),
    Protein = t(protein),
    Methylation = t(methylation)
)

Create and Train MOFA Model

Goal: Configure and train a MOFA2 model to discover shared and view-specific latent factors.

Approach: Set model and training options, then run variational inference to learn factor decomposition.

# Create MOFA object
mofa <- create_mofa(data_list)

# View data overview
plot_data_overview(mofa)

# Set model options
model_opts <- get_default_model_options(mofa)
model_opts$num_factors <- 15  # Number of factors to learn

# Set training options
train_opts <- get_default_training_options(mofa)
train_opts$convergence_mode <- 'slow'
train_opts$seed <- 42

# Prepare and train
mofa <- prepare_mofa(mofa, model_options = model_opts, training_options = train_opts)
mofa <- run_mofa(mofa, outfile = 'mofa_model.hdf5')

Analyze Factors

Goal: Quantify how much variance each factor explains per omics view and extract factor scores and loadings.

Approach: Plot variance decomposition and retrieve factor values (sample scores) and weights (feature loadings) as data frames.

# Variance explained per factor per view
plot_variance_explained(mofa, max_r2 = 15)
plot_variance_explained(mofa, plot_total = TRUE)[[2]]

# Factor values (sample scores)
factors <- get_factors(mofa, as.data.frame = TRUE)

# Factor weights (feature loadings)
weights <- get_weights(mofa, as.data.frame = TRUE)

Visualize Results

Goal: Generate publication-quality plots of factor values, feature weights, and factor correlations.

Approach: Use MOFA2 built-in plotting functions for scatter plots, heatmaps, and correlation matrices.

# Scatter plot of factor values
plot_factor(mofa, factor = 1, color_by = 'Group')

# Heatmap of factor weights
plot_weights(mofa, view = 'RNA', factor = 1, nfeatures = 20)

# Top features per factor
plot_top_weights(mofa, view = 'RNA', factor = 1, nfeatures = 10)

# Correlation between factors
plot_factor_cor(mofa)

Factor Interpretation

Goal: Identify biological pathways and gene sets associated with each MOFA factor.

Approach: Extract top-weighted features per factor and run gene set enrichment on factor weights.

# Extract top features for pathway analysis
top_rna_factor1 <- get_weights(mofa, views = 'RNA', factors = 1, as.data.frame = TRUE)
top_rna_factor1 <- top_rna_factor1[order(abs(top_rna_factor1$value), decreasing = TRUE), ]
gene_list <- head(top_rna_factor1$feature, 100)

# Gene set enrichment on factor weights
library(MOFA2)
enrichment <- run_enrichment(mofa, feature.sets = msigdb_genesets,
                              view = 'RNA', factors = 1:5)
plot_enrichment(enrichment, factor = 1, max.pathways = 15)

Add Sample Metadata

Goal: Annotate MOFA factors with clinical or experimental metadata for colored visualizations.

Approach: Load sample annotations and attach to the MOFA object, then plot factors colored by metadata variables.

# Load sample annotations
metadata <- read.csv('sample_metadata.csv', row.names = 1)

# Add to MOFA object
samples_metadata(mofa) <- metadata[samples_names(mofa)[[1]], ]

# Color by metadata
plot_factor(mofa, factor = 1, color_by = 'Condition')
plot_factors(mofa, factors = 1:3, color_by = 'Condition')

Multi-Group MOFA

Goal: Apply MOFA to batch- or group-structured multi-omics data and compare factor activity across groups.

Approach: Organize data as nested list by group, train multi-group MOFA, and visualize group-specific factor patterns.

# For batch/group-structured data
data_list_grouped <- list(
    group1 = list(RNA = rna_g1, Protein = prot_g1),
    group2 = list(RNA = rna_g2, Protein = prot_g2)
)

mofa_grouped <- create_mofa(data_list_grouped)
mofa_grouped <- prepare_mofa(mofa_grouped)
mofa_grouped <- run_mofa(mofa_grouped)

# Compare factor activity across groups
plot_factor(mofa_grouped, factor = 1, group_by = 'group', color_by = 'group')

MOFA+ for Single-Cell

Goal: Apply MOFA to single-cell multi-modal data (CITE-seq, Multiome) with stochastic inference for scalability.

Approach: Extract modality matrices from a Seurat object, create MOFA with stochastic training for large cell counts.

# For single-cell multi-omics (CITE-seq, Multiome)
library(Seurat)

# Extract modalities from Seurat object
rna_mat <- GetAssayData(seurat_obj, assay = 'RNA', layer = 'data')
adt_mat <- GetAssayData(seurat_obj, assay = 'ADT', layer = 'data')

# Create MOFA with single-cell settings
mofa_sc <- create_mofa(list(RNA = rna_mat, ADT = adt_mat))
model_opts <- get_default_model_options(mofa_sc)
model_opts$num_factors <- 10

# Use stochastic inference for large datasets
train_opts <- get_default_training_options(mofa_sc)
train_opts$stochastic <- TRUE

Export Results

Goal: Save MOFA factor scores, feature weights, and variance explained to CSV for downstream use.

Approach: Extract each result type as a data frame and write to disk.

# Save factor values
factors_df <- get_factors(mofa, as.data.frame = TRUE)
write.csv(factors_df, 'mofa_factors.csv', row.names = FALSE)

# Save weights
weights_df <- get_weights(mofa, as.data.frame = TRUE)
write.csv(weights_df, 'mofa_weights.csv', row.names = FALSE)

# Save variance explained
var_exp <- get_variance_explained(mofa)
write.csv(var_exp$r2_per_factor, 'mofa_variance_explained.csv')

Related Skills

• mixomics-analysis - Supervised multi-omics integration
• data-harmonization - Preprocess data before MOFA
• pathway-analysis/go-enrichment - Interpret MOFA factors
• single-cell/multimodal-integration - Single-cell multi-omics