nipalsMCIA: Flexible Multi-Block Dimensionality Reduction in R via Non-linear Iterative Partial Least Squares

Area of Science:

• Computational biology
• Bioinformatics
• Data science

Background:

• Multi-omics data analysis is crucial for bulk and single-cell studies.
• Unsupervised methods for clustering, visualization, and feature selection are essential.
• Existing Multiple Co-inertia Analysis (MCIA) implementations are not optimized for large datasets and lack out-of-sample embedding capabilities.

Purpose of the Study:

• To introduce nipalsMCIA, an efficient implementation of MCIA for multi-omics data.
• To address the limitations of current MCIA methods in terms of speed and scalability for single-cell data.
• To provide a tool for robust unsupervised analysis, including clustering, visualization, and feature selection.

Main Methods:

• Implemented MCIA using an extension of the Non-linear Iterative Partial Least Squares (NIPALS) algorithm.
• Developed a method that avoids eigendecompositions for improved speed and variance calculation.
• Enabled out-of-sample embedding for new data points.

Main Results:

• nipalsMCIA demonstrates significant speed-up compared to eigendecomposition-based MCIA methods for single-cell multi-omics data.
• The new implementation removes the need for eigendecomposition for variance explained calculations.
• Users can now perform out-of-sample embedding with nipalsMCIA.

Conclusions:

• nipalsMCIA offers an optimized and scalable solution for joint dimensionality reduction of multi-omics data.
• The package provides flexibility with various pre-processing and parameter options for downstream analysis.
• nipalsMCIA is available as a BioConductor package with comprehensive documentation.