bio-multi-omics-mofa-integration

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')