BiTSC

BiTSC identifies conserved gene co-clusters across two species by encoding orthology as a bipartite network with gene expression as node covariates to improve cross-species gene function inference.

Key Features:

• Bipartite Network Formulation: Encodes gene orthology as a bipartite network and incorporates gene expression data as node covariates to capture conservation and expression similarity.
• Unsupervised Learning Techniques: Combines kernel enhancement, bipartite spectral clustering, consensus clustering, tight clustering, and hierarchical clustering and does not mandate inclusion of all genes into clusters.
• Distribution-Free: Operates without relying on distributional assumptions about the data.
• General Applicability: Applies as a general algorithm for identifying tight node co-clusters in any bipartite network with node covariates.

Scientific Applications:

• Cross-Species Gene Co-Clustering: Identifies conserved co-clusters between species such as Drosophila melanogaster and Caenorhabditis elegans to aid prediction of unknown gene functions.
• Validation and Biological Significance: Has been evaluated by simulation studies demonstrating accuracy and robustness and used to validate co-clusters and verify their biological significance.

Methodology:

Constructs a bipartite network from orthologous gene relationships with gene expression as node covariates, applies kernel enhancement and bipartite spectral clustering, and refines clusters via consensus, tight, and hierarchical clustering.