LDA-DC

LDA-DC performs probabilistic double clustering to jointly stratify patients and cell populations from high-dimensional omics datasets such as flow cytometry and microbiota profiles.

Key Features:

• Double clustering: Simultaneously identifies clusters of patients and related clusters of cells within a unified model.
• LDA extension: Implements an adaptation of Latent Dirichlet Allocation (LDA) to model biological observations rather than text topics.
• Probabilistic generative model: Uses a generative probabilistic framework to represent latent patient- and cell-level structure.
• Joint phenotype extraction: Extracts meaningful phenotypes from both patient-level and cell-level data concurrently.
• Handles cohort variability: Designed to account for cohort variability and patient heterogeneity in high-dimensional datasets.
• Validation on synthetic data: Validated using artificial datasets to assess performance on heterogeneous, large-scale data.
• Application to cytometry and microbiota data: Applied to flow cytometry and microbiota datasets for real-world stratification tasks.
• Computational efficiency: Described as computationally efficient for handling large omics datasets.

Scientific Applications:

• Patient stratification: Stratifies patient cohorts into distinct subgroups based on multi-level omics signals.
• Diagnostics: Supports diagnostic inference by linking patient clusters to characteristic cell-population profiles.
• Identification of disease-associated cell populations: Discovers cell clusters associated with specific patient conditions.
• Analysis of flow cytometry and microbiota datasets: Enables joint analysis of cytometry and microbiota data for integrative studies.
• Pre-clinical research and clinical applications: Facilitates investigation of disease mechanisms and patient-specific conditions in pre-clinical and clinical contexts.

Methodology:

An extension of Latent Dirichlet Allocation (LDA) is used as a probabilistic generative model for double clustering, validated on artificial datasets and applied to flow cytometry and microbiota data to jointly infer patient clusters and associated cell-population clusters.