DCMD

DCMD applies distance-based classification using mixture distributions to model sparsity and uncertainty in next-generation sequencing microbiome count data for improved classification.

Key Features:

• Mixture distribution estimation: Estimates a mixture distribution from sample count data to model population-level heterogeneity.
• Conditional observation representation: Represents each observation as a conditional distribution based on observed counts and the estimated mixture for downstream analysis.
• Distance-based metrics: Implements two specifically developed distance metrics tailored to sparse, heterogeneous microbiome count data.
• Integration with classifiers: Uses the derived distance metrics within k-means clustering and k-nearest neighbors (k-NN) classification frameworks.
• Sparsity and uncertainty modeling: Explicitly models uncertainty in sparse counts typical of next-generation sequencing microbiome data.
• Empirical evaluation: Evaluated on simulated datasets and real human microbiome study data and compared against conventional distance-based classifiers and other machine learning approaches.

Scientific Applications:

• Microbiome sample classification: Classifying microbiome samples derived from next-generation sequencing count data.
• Clustering of microbial communities: Performing k-means clustering using DCMD-derived distances to capture community structure.
• k-NN classification: Applying k-nearest neighbors classification with DCMD distance metrics for sample labeling.
• Study of microbiome–health relationships: Analyzing connections between microbial community composition and human health outcomes using sparse count data.
• Method benchmarking: Benchmarking and comparative evaluation of classifiers on simulated and real human microbiome datasets.

Methodology:

Estimate a mixture distribution from sample counts; represent each observation as a conditional distribution given observed counts and the estimated mixture; compute two specialized distance metrics and apply them in distance-based classification methods including k-means and k-nearest neighbors; evaluate performance on simulated and human microbiome study data.