rama

rama estimates gene expression intensities from cDNA microarray replicates using a Bayesian hierarchical model to provide robust intensity estimates in the presence of outliers and nonconstant variance.

Key Features:

• Bayesian Hierarchical Model: Provides robust estimation of gene intensities by modeling multiple levels of variation inherent in cDNA microarray experiments.
• Outlier Modeling (t-distribution): Explicitly models outliers using a t-distribution to reduce their influence on intensity estimates.
• Design Effects: Accounts for design-related variations across arrays and replicates in the statistical model.
• Normalization and Transformation: Includes methods for normalization and transformation of cDNA microarray data to improve consistency across datasets.
• Nonconstant Variance (Heteroscedasticity): Addresses heteroscedasticity by allowing variance to vary across levels of the data.
• Parameter Estimation (MCMC): Uses Markov chain Monte Carlo (MCMC) techniques for parameter estimation and uncertainty quantification.
• Automatic Quality Control: Identifies potential outliers at the replicate, array, and gene levels to support data quality assessment.
• Background Subtraction Handling: Provides an approach that permits background subtraction without substantially increasing variability in intensity estimates.

Scientific Applications:

• Gene Filtering: Facilitates filtering of genes based on robust intensity estimates and quality metrics.
• Heteroscedasticity Management: Enables analysis that accounts for nonconstant variance across probes or conditions.
• Quality Control: Detects outliers at replicate, array, and gene levels to improve reliability of microarray results.
• Comparative Evaluation: Has been applied to public gene expression datasets and shown lower between-replicate variability compared with ANOVA-normalized log ratios, median log ratio, and Dixon's test-based outlier removal.

Methodology:

Uses a Bayesian hierarchical model with outliers modeled by a t-distribution, accounts for design effects, applies normalization and transformation, supports background subtraction, and estimates parameters via Markov chain Monte Carlo (MCMC).