Assessment of statistical methods used in library-based approaches to microbial source tracking

Area of Science:

• Environmental Microbiology
• Bioinformatics
• Statistical Modeling

Background:

• Microbial source tracking (MST) relies on statistical methods for accurate source identification.
• A lack of consensus exists regarding the most effective statistical approaches for MST.

Purpose of the Study:

• To evaluate the effectiveness of various statistical pattern-matching algorithms in microbial source tracking.
• To compare the performance of discriminant analysis, nearest neighbor, maximum similarity, and average similarity methods.

Main Methods:

• Six independent libraries were created from human, seagull, cow, and dog fecal samples.
• rep-PCR and antibiotic resistance analysis (ARA) techniques were used to construct libraries.
• Various statistical methods and threshold criteria were applied to assess classification success.

Main Results:

• Significant variability was observed in the correct classification rates among different statistical methods.
• No single statistical method consistently outperformed others in source identification.
• Threshold criteria did not reliably improve classification accuracy and often reduced effective sample size.

Conclusions:

• The choice of statistical method significantly impacts microbial source tracking outcomes.
• Further research is needed to establish standardized and superior statistical approaches for MST.
• Recommendations are provided for selecting appropriate statistical analyses for MST data.