fabia

fabia identifies biclusters in transcriptomic gene expression data by using a generative multiplicative model to extract subsets of genes and samples with coherent expression patterns.

Key Features:

• Factor analysis: Uses a factor analysis framework to detect biclusters as latent factors linking genes and samples.
• Generative multiplicative model: Employs a multiplicative generative model to capture linear dependencies between gene expressions and experimental conditions.
• Heavy-tailed distribution handling: Accommodates heavy-tailed distributions commonly observed in real-world transcriptomic datasets.
• Bicluster identification: Identifies subsets of genes and samples exhibiting similar expression patterns (biclusters).
• Information-content ranking: Ranks detected biclusters based on information content to prioritize biologically relevant clusters.
• Model selection and Bayesian techniques: Integrates model selection methods and Bayesian techniques within its generative framework.
• Empirical performance: Demonstrated superior performance in simulations (100 datasets with known biclusters, outperforming 11 competing methods) and top-ranking results on three microarray datasets with pre-identified subclusters.
• Implementation: Implemented in R.

Scientific Applications:

• Biclustering of transcriptomic/microarray data: Discovery of co-expressed gene modules across subsets of samples in transcriptomic and microarray studies.
• Sample subgroup identification: Detection and characterization of sample subclusters or experimental-condition-specific expression patterns.
• Prioritization for biological interpretation: Prioritization of biclusters for downstream analysis to infer underlying biological processes or disease mechanisms.

Methodology:

Uses factor analysis within a multiplicative generative model, ranks biclusters by information content, and applies model selection and Bayesian techniques while accounting for linear dependencies and heavy-tailed expression distributions.