Predicting fecal sources in waters with diverse pollution loads using general and molecular host-specific indicators...

Area of Science:

• Environmental microbiology
• Water quality assessment
• Machine learning applications

Background:

• Fecal contamination in water poses significant public health risks.
• Accurate identification of contamination sources is crucial for effective water management.
• Traditional microbial source tracking (MST) methods face challenges with low-level pollution.

Purpose of the Study:

• To evaluate the efficacy of machine learning software (Ichnaea) in identifying fecal contamination sources in water.
• To assess the performance of various MST methods under different contamination scenarios.
• To develop predictive models for low-level fecal pollution detection.

Main Methods:

• Application of machine learning software (Ichnaea) for predictive modeling.
• Utilized multiple MST methods including host-specific phages, mitochondrial DNA, and bacterial markers.
• Employed general indicators like Escherichia coli and enterococci for comparison.

Main Results:

• Most MST methods accurately identified contamination at point source and moderate levels.
• Some indicators became undetectable at low contamination levels (< 3 log10 CFU E. coli/100 ml).
• Ichnaea software successfully generated models for low-level pollution using optimized MST methods.

Conclusions:

• Inductive machine learning offers a promising advancement for MST studies.
• The Ichnaea software demonstrates potential for accurate fecal contamination source identification.
• Method selection for MST should be scenario-dependent for optimal results.