Covariate-dependent Hierarchical Dirichlet Processes

2025-12-31 · Source: JMLR · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics, Computational Biology · Depth: Expert, quick

Summary

A new hierarchical Bayesian approach, Covariate-dependent Hierarchical Dirichlet Processes (CD-HDP), is proposed for density estimation and cluster identification across related groups. This method integrates covariate information and combines hierarchical Dirichlet processes with dependent Dirichlet processes, offering flexibility for multiple and mixed covariate types via kernel functions and various output types through component-specific likelihoods. The CD-HDP model enhances the ability to discern relationships between covariates and clusters, effectively borrowing information and quantifying group differences. Posterior inference is performed using a Markov chain Monte Carlo algorithm, facilitated by a data augmentation trick to handle intractable normalized weights. The model's efficacy is demonstrated on simulated data, single-cell RNA sequencing (scRNA-seq) data, and calcium imaging data, revealing additional cell subgroups and interpretable neural activity clusters, respectively.

Key takeaway

For research scientists working with complex biological data like scRNA-seq or calcium imaging, you should consider applying Covariate-dependent Hierarchical Dirichlet Processes (CD-HDP). This method can reveal more nuanced subgroups and interpretable clusters by incorporating covariate information, potentially leading to deeper biological insights than traditional hierarchical models. Evaluate its performance against existing methods for improved density estimation and cluster identification.

Key insights

CD-HDP integrates covariates into hierarchical Bayesian nonparametrics for flexible density estimation and cluster identification.

Principles

• Integrate covariates for enhanced cluster discovery.
• Combine HDP with DDP for model flexibility.
• Borrow information across groups effectively.

Method

The CD-HDP model uses a data augmentation trick to handle intractable normalized weights, enabling posterior inference via a Markov chain Monte Carlo algorithm for density estimation and cluster identification.

In practice

• Apply to scRNA-seq for cell subgroup discovery.
• Use for calcium imaging to identify neural activity clusters.

Topics

• Hierarchical Dirichlet Processes
• Bayesian Nonparametrics
• Density Estimation
• Cluster Analysis
• Single-cell RNA sequencing

Best for: Research Scientist, AI Researcher, AI Scientist, Data Scientist

Related on AIssential

• Bayesian inference of haematopoietic stem/progenitor cell differentiation phenotypic manifolds and their bifurcation points using Gaussian processes and Gibbs sampling — BAGEL provides a Bayesian framework for continuous cell differentiation modeling and multi-dataset visualization.
• Hierarchical Causal Models — Hierarchical Causal Models extend structural causal models to enable causal inference in nested data.
• Estimating Mutual Information between Time Series and Temporal Event Sequences Across Diverse Analysis Tasks — A nonparametric mutual information estimator quantifies dependence between heterogeneous time series and event sequences without transformation or discretization.
• Kernel of Partition Paths: A Unified Representation for Tree Ensembles — KPP unifies prediction, attribution, robustness, and generalization for tree ensembles via a node-indexed, path-isometric feature map.
• Scalable Model-Based Clustering with Sequential Monte Carlo — Decomposing online clustering problems into independent subproblems enables scalable Sequential Monte Carlo inference.
• Decorrelated Local Linear Estimator: Inference for Non-linear Effects in High-dimensional Additive Models — A new decorrelated local linear estimator enables robust inference for non-linear effects in high-dimensional additive models.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by JMLR.