PLA

PLA performs piecewise-constant and low-rank matrix approximation to identify recurrent copy number variations (CNVs) across multi-sample array-based comparative genomic hybridization (aCGH) profiles.

Key Features:

• Matrix representation: Represents multi-sample aCGH data as a matrix in which recurrent CNVs manifest as a low-rank, piecewise-constant structure.
• Matrix recovery formulation: Transforms CNV identification into a matrix recovery problem aiming for an optimal piecewise-constant and low-rank approximation.
• Convex optimization: Employs a convex formulation for the matrix recovery problem.
• Global solution algorithm: Uses an efficient algorithm designed to globally solve the convex matrix recovery formulation.
• Multi-sample focus: Detects recurrent CNVs across multiple samples rather than performing single-sample analysis.
• Validation: Performance validated on synthesized datasets and two breast cancer datasets.

Scientific Applications:

• Recurrent CNV detection: Identification of recurrent copy number variations across multi-sample aCGH datasets.
• Cancer genomics: Analysis of recurrent genomic alterations in cancer, demonstrated on breast cancer datasets.
• Comparative multi-sample analysis: Comparative reconstruction of CNV patterns across samples for studies of disease mechanisms.
• Method benchmarking: Provides a convex matrix-recovery framework usable for benchmarking CNV reconstruction on synthesized data.

Methodology:

Represents input aCGH profiles as a matrix where recurrent CNVs correspond to low-rank components, formulates identification as a convex matrix recovery problem seeking a piecewise-constant and low-rank approximation, and solves it with an efficient algorithm designed to obtain a global solution.