UniCor and UniCorP: a novel metric and hierarchical feature selection algorithm for microbial community analysis

Area of Science:

• Life Sciences
• Bioinformatics
• Data Analysis

Background:

• Life sciences generate big data, posing analysis challenges.
• Microbiome datasets have complex hierarchical structures often underutilized.
• Existing methods may not fully leverage feature hierarchies for analysis.

Purpose of the Study:

• Introduce UniCor, a novel metric for identifying uniquely correlated entities (UNICORNs).
• Develop UniCorP, a propagation algorithm to exploit dataset hierarchies.
• Enhance feature reduction and predictive performance in quantitative datasets.

Main Methods:

• UniCor metric combines feature uniqueness and correlation to a target variable.
• UniCorP algorithm propagates features based on UniCor scores, utilizing dataset hierarchies.
• Random Forest Regressions used for cross-validation with bacterial community data.

Main Results:

• UniCorP successfully reduced features and improved predictive performance in microbiome datasets.
• The metric outperformed control trials across taxonomic aggregation levels.
• UniCorP demonstrated stable and competitive performance compared to existing methods.

Conclusions:

• UniCor and UniCorP offer a simple, adaptable framework for feature reduction and analysis.
• The approach effectively utilizes hierarchical structures in big data, like microbiome data.
• This method enhances predictive modeling while substantially reducing the feature space.