EDCluster

EDCluster constructs empirical stationary distribution mixture models to model compositional heterogeneity across amino acid sites and improve phylogenetic inference.

Key Features:

• Cluster analysis with coordinate transformations: Performs cluster analysis incorporating specific coordinate transformations to detect specialized amino acid distributions from curated databases or directly from sequence alignments.
• Maximum likelihood estimation: Uses maximum likelihood estimation to derive empirical stationary distribution mixture models.
• High component scalability: Estimates mixture models with up to 4,096 components for large-scale analyses.
• Application to curated databases: Generates universal distribution mixture (UDM) models by applying methods to HOGENOM and HSSP databases.
• Improved phylogenetic performance: Reduces systematic errors such as long-branch attraction and shows improvements relative to C10–C60 models.
• Software compatibility: Provides models compatible with IQ-TREE, PhyloBayes, and RevBayes.

Scientific Applications:

• Modeling amino acid site heterogeneity: Enables construction of empirical distribution mixtures to capture site-specific compositional variability in protein alignments.
• Phylogenetic inference improvement: Enhances accuracy of phylogenetic analyses by mitigating biases (e.g., long-branch attraction) arising from compositional heterogeneity.

Methodology:

Performs cluster analysis with specific coordinate transformations and applies maximum likelihood estimation to derive empirical stationary distribution mixture models (up to 4,096 components), including construction of UDM models from HOGENOM and HSSP.