codaloss

codaloss infers direct interaction networks among microbial species from compositional microbiome data, enabling estimation of direct microbial interactions from 16S rRNA gene sequencing and whole microbiome sequencing datasets.

Key Features:

• Novel Loss Function: Introduces a loss function specifically tailored to the compositional nature of microbiome data to facilitate estimation of direct interactions under sparsity assumptions.
• Sparse Solution Optimization: Implements an alternating direction optimization algorithm to obtain sparse network solutions.
• Reduced Assumptions: Operates with fewer assumptions about microbial network structure to allow more flexible modeling of complex interactions.
• Performance Evaluation: Demonstrated superior performance in simulation studies and analyses of real microbiome datasets compared to other state-of-the-art methods.

Scientific Applications:

• Microbiome network inference: Estimate direct species-species interactions from compositional datasets such as 16S rRNA gene sequencing and whole microbiome sequencing.
• Community ecology and regulation: Investigate mechanisms regulating microbial community structure and interactions.
• Health, disease, and ecological studies: Support analyses of host-microbe interactions and microbial contributions to health, disease, and ecological processes.

Methodology:

Uses a novel loss function tailored to compositional data, implements an alternating direction optimization algorithm to obtain sparse solutions, and relies on sparsity assumptions while minimizing assumptions about network structure.