K2Taxonomer

K2Taxonomer identifies nested, taxonomy-like subgroups in high-throughput -omics datasets to enable multi-resolution characterization of molecular taxonomies in bulk and single-cell transcriptomics.

Key Features:

• R implementation: Provided as an R package for analysis of -omics datasets.
• Unsupervised nested subgroup discovery: Performs unsupervised learning to detect nested, taxonomy-like sample subgroups.
• Top-down recursive partitioning: Uses a top-down recursive partitioning framework to generate hierarchical partitions.
• Ensemble learning: Leverages ensemble learning techniques to enhance robustness of subgroup identification.
• Data modality support: Supports analysis of both bulk and single-cell transcriptomics data.
• Gene- and pathway-level annotation: Annotates identified subgroups with gene- and pathway-level analyses for biological interpretation.
• Hierarchical structure recovery: Recovers hierarchical structures that resemble taxonomies within molecular datasets.
• Dataset types: Applicable to simulated datasets and real-world human tissue transcriptomics data.

Scientific Applications:

• Exploratory discovery: Enables exploration of sample relationships when groupings are not predefined, allowing data-driven subgroup discovery.
• Bulk and single-cell transcriptomics analysis: Characterizes multi-resolution molecular structure in both bulk and single-cell expression studies.
• Breast cancer TIL single-cell profiling: Used to analyze tumor-infiltrating lymphocyte single-cell profiles in breast cancer, identifying co-expression patterns of translational machinery genes across T cell subtypes.
• Prognostic association discovery: Identified co-expression patterns in T cells that were associated with improved prognosis in bulk breast cancer expression data.
• Hypothesis generation and validation support: Provides subgroup annotations that guide hypothesis generation and downstream experimental validation.

Methodology:

Performs unsupervised learning using a top-down recursive partitioning framework combined with ensemble learning, with subgroup annotation via gene- and pathway-level analyses.