Metagenomic Geolocation Prediction Using an Adaptive Ensemble Classifier.

Area of Science:

• Environmental microbiology
• Bioinformatics
• Computational biology

Background:

• Urban environments harbor distinct microbial communities.
• Geographic location can be inferred from microbial signatures in environmental samples.

Purpose of the Study:

• To predict the geographic origin of unknown microbiome samples using urban microbial signatures.
• To evaluate the performance of various classification algorithms for metagenomic sample source prediction.

Main Methods:

• Standard bioinformatics pipelines were used to pre-process raw metagenomic data.
• Taxonomy-dependent and taxonomy-free approaches were employed to train component classifiers.
• Class weighting and optimal oversampling techniques addressed data imbalance.
• An ensemble classifier was developed by combining multiple component classifiers.

Main Results:

• Component classifiers showed varied performance in predicting sample origins.
• The ensemble classifier consistently delivered optimal prediction performance.
• The ensemble approach matched the performance of the best-performing individual classifier.

Conclusions:

• Relying on a single classification algorithm for metagenomic sample source prediction is unreliable.
• Ensemble methods, by integrating multiple classifiers, enhance prediction accuracy and robustness.
• Microbiome analysis of urban environments offers a viable strategy for geographic source attribution.