PIntMF

PIntMF performs penalized matrix factorization to integrate multi-omics data, cluster samples, and select relevant omics variables for biological interpretation.

Key Features:

• Sparsity and Positivity Constraints: Employs Lasso penalization to induce sparsity in both the variable and individual matrices while enforcing positivity constraints to maintain interpretability.
• Equality Constraints for Normalization: Implements equality constraints to standardize inferred coefficients and normalize factorization results.
• Automatic Tuning with glmnet: Uses the glmnet package to automatically adjust sparsity parameters for penalization.
• Criteria for Latent Variables Selection: Provides three criteria to assist in determining the appropriate number of latent variables.

Scientific Applications:

• Benchmarking: Compared with other integrative methods and feature selection techniques on synthetic and real datasets.
• Simulation Studies: Identifies meaningful clusters and relevant variables in simulated data with correlated or uncorrelated structures.
• Diet Dataset Analysis: Applied to a diet-related real dataset to identify interpretable clusters corresponding with available clinical data.
• Cancer Dataset Analysis: Applied to a cancer real dataset to identify interpretable clusters corresponding with available clinical data.

Methodology:

Penalized matrix factorization with Lasso penalization, positivity and equality constraints for normalization, automatic tuning of sparsity parameters via glmnet, and three criteria for selecting the number of latent variables; outputs include sample clusters and selected omics variables.