From microbial diversity to functional potential using dimensionality reduction.

Area of Science:

• Microbial Ecology
• Oceanography
• Machine Learning

Background:

• High-dimensional microbial 'omics data present challenges for ecological analysis.
• Machine learning (ML) offers powerful tools for dimensionality reduction and feature discovery.
• Understanding Arctic microbial community dynamics is crucial for biogeochemical processes.

Purpose of the Study:

• Compare Self Organizing Map (SOM) to Principal Coordinate Analysis (PCoA) and Weighted Gene Correlation Network Analysis (WGCNA) for microbial data.
• Identify and characterize functionally distinct microbial ecotypes in the Arctic Ocean.
• Assess the utility of ML for analyzing microbial succession and diversity.

Main Methods:

• Utilized 16S rRNA gene amplicon sequencing data from the MOSAiC expedition (2019-2020).
• Applied SOM, PCoA, and WGCNA for dimensionality reduction and clustering.
• Integrated k-means clustering outputs with metagenomic data to define ecotypes.

Main Results:

• SOM outperformed PCoA in representing seasonal microbial transitions.
• SOM identified more metabolically distinct functional groups than PCoA.
• Four distinct microbial ecotypes were identified, driven by seasonality, water mass, and substrate turnover.

Conclusions:

• ML dimensionality reduction, particularly SOM, is effective for analyzing microbial 'omics data.
• Seasonal succession plays a key role in Arctic microbial functional diversity.
• ML-derived ecotypes can enhance biogeochemical models for the Arctic Ocean.