rCUR: an R package for CUR matrix decomposition.

Area of Science:

• Computational Biology
• Data Science
• Bioinformatics

Background:

• Singular Value Decomposition (SVD) is a common dimensionality reduction technique for large datasets like microarrays.
• SVD-derived singular vectors are difficult to interpret in the original data domain.
• CUR matrix decompositions offer an interpretable alternative by using actual data columns/rows.

Purpose of the Study:

• To introduce an R-package, rCUR, for performing CUR matrix decompositions.
• To enable interactive and exploratory CUR-based analysis on large-scale, high-throughput data.
• To demonstrate the utility of CUR decompositions in maintaining data trends and classification accuracy while reducing data size.

Main Methods:

• Implementation of CUR matrix decomposition algorithms in an open-source R-package (rCUR).
• Utilizing statistical leverage scores for identifying informative data points.
• Applying CUR-based techniques to high-throughput biological data.

Main Results:

• The rCUR package provides functions for CUR-based matrix decomposition in R.
• CUR-based analysis can significantly reduce the number of data probes (features).
• Maintained major data trends and classification accuracy were observed in examples.

Conclusions:

• rCUR enables CUR-based matrix decomposition within the R environment.
• Facilitates differential expression analysis and discriminant gene selection in gene expression studies.
• Leverage scores identify key data points for expressing remaining data, aiding in outlier detection and interpretation.