MGMM

MGMM estimates Gaussian mixture model parameters and performs clustering and classification while directly handling missing values in omics datasets.

Key Features:

• Missing Data Accommodation: Integrates missing values directly into GMM fitting without requiring pre-imputation or complete-case analysis and imposes no restrictions on the covariance matrix form.
• Enhanced Parameter Estimation: Produces more reliable parameter estimates by avoiding biases and instability associated with pre-imputation or complete-case approaches.
• Robust Clustering Performance: Recovers true cluster assignments more accurately than GMM implementations that rely on imputation or inadequately handle missing data, based on evaluations using real and simulated omics datasets.
• Uncertainty Assessment: Quantifies cluster assignment uncertainty and remains effective when underlying data distributions deviate from Gaussian mixture assumptions.
• Versatility Across Data Sets and Missingness Rates: Demonstrates superior recovery of true clusters across diverse datasets and varying levels of missing data in benchmark evaluations.

Scientific Applications:

• Clustering and density estimation: Application to genomics, proteomics, and other 'omics datasets for cluster discovery and density modeling.
• Integrative analysis: Integration of multiple, disparate datasets that contain missing values without pre-imputation.
• Classification with uncertainty quantification: Assigning samples to mixture components while providing measures of assignment uncertainty, including under model misspecification.

Methodology:

Fits Gaussian Mixture Models by incorporating missing data directly into model fitting (avoiding pre-imputation and complete-case analysis), permits unrestricted covariance matrices, and produces parameter estimates, classifications, and cluster-assignment uncertainty measures.