UniCoracle: automated hierarchical feature selection via bottom-up propagation and top-down skimming using the...

Area of Science:

• Microbiome bioinformatics
• Computational biology
• Machine learning applications in ecology

Background:

• Microbiome data analysis is computationally intensive and complex.
• Existing machine learning (ML) frameworks like Coracle aid in identifying associations but can be further optimized.
• Automated feature aggregation algorithms like UniCorP have been developed to enhance analysis.

Purpose of the Study:

• To introduce UniCoracle, a novel automated analytical framework for microbiome data.
• To integrate UniCorP's bottom-up approach with a new top-down skimming (TDS) strategy within the Coracle ML framework.
• To leverage the taxonomic structure of microbiome data for improved predictive stability and biological insight.

Main Methods:

• UniCoracle integrates UniCorP's bottom-up feature aggregation with a top-down skimming (TDS) strategy.
• The framework is implemented using the Coracle machine learning (ML) framework.
• It processes microbiome data, such as ASVs from 16S rRNA gene sequencing, to identify taxonomic associations.

Main Results:

• UniCoracle demonstrates competitive or improved predictive performance compared to non-hierarchical Coracle and TDS-based Coracle methods.
• The framework achieves enhanced predictive accuracy.
• UniCoracle effectively identifies specific taxonomic features associated with continuous target variables.

Conclusions:

• UniCoracle offers a streamlined, user-friendly framework for microbiome data analysis and hypothesis generation.
• It provides control over feature set size and runtime, reducing computational demands.
• The framework successfully identifies biologically meaningful taxonomic associations at specific hierarchical levels.