Hierarchical classification of microorganisms based on high-dimensional phenotypic data.

Area of Science:

• Microbiology
• Spectroscopy
• Bioinformatics

Background:

• Microorganism classification is challenging due to complex phylogenetic taxonomy.
• High-dimensional phenotyping methods like FTIR spectroscopy generate complex data.
• Hierarchical structures can aid classification by integrating phylogenetic or phenotypic data.

Purpose of the Study:

• To compare different hierarchical classification approaches for microorganisms using high-dimensional phenotypic data.
• To evaluate the effectiveness of various classification algorithms on FTIR spectroscopic data of molds.
• To introduce and assess the random forest approach for microbial classification via FTIR spectroscopy.

Main Methods:

• Utilized a collection of 19 mold species (filamentous fungi).
• Employed hierarchical cluster analysis to establish classification trees.
• Compared classification algorithms: artificial neural networks (ANN), partial least-squared discriminant analysis, and random forest (RF).

Main Results:

• Artificial neural networks (ANN) and random forest (RF) demonstrated superior performance compared to other methods.
• These top-performing methods did not require a predefined hierarchical structure.
• The random forest (RF) approach was applied for microorganism classification using FTIR spectroscopy for the first time.

Conclusions:

• Non-hierarchical machine learning methods, specifically ANN and RF, are highly effective for microorganism classification based on FTIR data.
• The random forest algorithm presents a promising and novel tool for microbial identification using spectroscopic phenotyping.
• Future research should explore the broader applicability of RF in microbial classification.