plw

plw performs probe-level differential expression analysis of Affymetrix and other multiple-probe microarray platforms using locally moderated weighted median-t (PLW) and locally moderated weighted-t (LMW) statistics to identify differentially expressed genes.

Key Features:

• Empirical Bayes Framework: Implements an empirical Bayes model that adjusts sample variances toward a global estimate while accounting for the dependency between variability and intensity levels.
• Intensity-Level Dependency: Models intensity-dependent variance to reduce intensity-level dependent false positive rates common in microarray data.
• Global Covariance Matrix: Estimates a global covariance matrix to accommodate differing variances between arrays and array-to-array correlations.
• Probe-Level Analysis: Performs inference at the individual probe level for Affymetrix-type arrays and summarizes probe-level results into a single probe-set score.
• Statistical Tests: Uses locally moderated weighted median-t statistics (PLW) and locally moderated weighted-t statistics (LMW) and compares performance to moderated t-tests.
• Benchmarking and Validation: Validated against 14 existing methods on five spike-in datasets, with PLW ranking regulated genes most accurately in four of five datasets on RMA and GCRMA, and LMW outperforming moderated t-tests across RMA, GCRMA, and MAS5 expression indexes.

Scientific Applications:

• Differential Expression Detection: Identifying differentially expressed genes in microarray experiments.
• Cancer Genomics: Detecting gene expression changes relevant to cancer genomics studies.
• Developmental Biology: Analyzing expression patterns in developmental biology research.
• Robust Microarray Analysis: Analyses of Affymetrix and other multiple-probe microarray datasets processed with RMA, GCRMA, or MAS5 that require intensity-dependent variance modeling.

Methodology:

Empirical Bayes variance moderation; modeling intensity-dependent variability; estimation of a global covariance matrix for array variances and correlations; probe-level inference with weighted median-t statistics for PLW and locally moderated weighted-t for LMW; evaluation on spike-in datasets with RMA, GCRMA, and MAS5-processed data.