coseq

coseq performs clustering of compositional high-throughput RNA sequencing feature profiles to identify groups of co-expressed genes and other compositional patterns.

Key Features:

• Clustering of Expression Profiles: Clusters feature profiles derived from high-throughput RNA sequencing, accommodating features with zero or near-zero counts across samples.
• Adapted Data Transformations: Implements Centered Log Ratio (CLR) and Log Centered Log Ratio (LCL) transformations to address the compositional nature of the data.
• Clustering Algorithms: Supports mixture models and K-means for grouping transformed feature profiles.
• Model Selection Criteria: Employs a non-asymptotic penalized criterion with the penalty calibrated by slope heuristics to determine the optimal number of clusters.

Scientific Applications:

• Gene Expression Analysis: Identifies groups of co-expressed genes from RNA sequencing data, including cases where genes are silent under specific experimental conditions.
• Bicycle Sharing System Patterns: Detects usage patterns in the Velib' bicycle sharing system in Paris as an example of compositional data analysis beyond genomics.

Methodology:

Applies CLR and LCL transformations to compositional count data, clusters transformed profiles using mixture models or K-means, and selects the number of clusters via a non-asymptotic penalized criterion with penalty calibrated by slope heuristics.