Identification of human pathogens in soil by virulence gene-based machine learning method.

Area of Science:

• Environmental microbiology
• Bioinformatics
• Machine learning applications in pathogen detection

Background:

• Soils harbor human pathogenic bacteria, posing a public health risk.
• Metagenomics offers pathogen identification but faces limitations like time-consuming assembly and reliance on reference databases.
• Existing methods may miss novel or uncharacterized pathogens.

Purpose of the Study:

• To develop and validate a novel machine learning method for identifying human pathogenic bacteria in soil metagenomes.
• To leverage virulence factors (VFs) for improved pathogen detection accuracy and scope.
• To assess the abundance and distribution of soil pathogens across different land types in China.

Main Methods:

• Developed a virulence factor (VF) based K-Nearest Neighbors (VF-KNN) machine learning model.
• Trained the model on VF features of pathogenic and non-pathogenic bacteria.
• Validated the model using soil metagenomic data and isolated pathogenic strains, assessing accuracy and genome coverage.

Main Results:

• VF-KNN achieved high performance in soil pathogen identification (AUC: 0.95, Accuracy: 0.85), validated with 0.95 accuracy on isolated strains.
• The model demonstrated >0.90 prediction accuracy for top soil pathogens at 0.4X-1.0X genome coverage.
• VF-KNN identified 28% more potential pathogenic species than conventional methods, including newly reported ones like *Mycolicibacterium cosmeticum*.

Conclusions:

• The VF-KNN method offers a robust and efficient approach for identifying human pathogenic bacteria in soil metagenomes.
• This method expands the detection of potential pathogens beyond predefined lists and reference genomes.
• Soil pathogens are more abundant and diverse in agricultural lands, with significant presence in eastern China's topsoils.