fuseMLR

fuseMLR enables integrative prediction modeling of multi-omics data through modality-specific variable selection and prediction modeling followed by aggregation into a final meta-model.

Key Features:

• Late Integration Framework: Analyzes each data modality separately to obtain modality-specific predictions before aggregation.
• Modality-Specific Variable Selection: Performs variable selection independently for each data modality.
• Flexible ML Algorithm Application: Allows different machine learning algorithms to be applied for each modality.
• Automated Aggregation: Automatically aggregates modality-specific predictions once training is completed, either by training a machine learning meta-model or by computing the weighted mean of modality-specific predictions.
• Systematic and Reproducible Workflow: Implements a structured approach to late integration predictive modeling.

Scientific Applications:

• Multi-Omics Predictive Modeling: Integrates different types or modalities of molecular data measured from the same individuals for predictive modeling.
• Cancer Genomics: Applicable to multi-omics datasets such as those from The Cancer Genome Atlas.
• Partially Overlapping Patient Data: Addresses scenarios where patient data only partially overlap across modalities, avoiding the need to reduce datasets or impute missing values.

Methodology:

fuseMLR implements late integration prediction modeling in R, where each data modality is analyzed separately with modality-specific variable selection and machine learning algorithms, followed by aggregation of predictions into a meta-model through either machine learning or weighted mean computation.