SourceFinder

SourceFinder classifies chromosomal, plasmid, and bacteriophage sequences from bacterial genome assemblies using k-mer distributions and a random forest machine-learning classifier to identify sequence origin.

Key Features:

• Three-Class Classification: Distinguishes chromosomal, plasmid, and bacteriophage sequences using k-mer distribution analysis.
• Random Forest Classifier: Utilizes a random forest algorithm as the core predictive model.
• Training Dataset: Trained on 23,211 sequences from bacterial chromosomes, plasmids, and bacteriophages across hundreds of species.
• Fragmentation Strategy: Complete sequences were fragmented into 5,000-nucleotide segments to represent sequence plasticity and incomplete assemblies.
• K-mer Subdivision: Sequence fragments were subdivided into k-mers to capture short-subsequence patterns used for classification.
• Performance (AUC): Demonstrates robust performance with a minimum area under the receiver operating characteristic curve (AUC) of 0.939, including on simulated metagenomic scaffolds.
• Adaptability to Incomplete Data: Training on subsampled and fragmented sequences enables handling of incomplete genomic assemblies.

Scientific Applications:

• Extra-chromosomal element identification: Detects plasmids and bacteriophages that encode antimicrobial resistance, metal resistance, and virulence genes.
• Gene dissemination analysis: Assesses the distribution and dissemination of accessory genes across microbial communities.
• Horizontal gene transfer and evolutionary studies: Supports analysis of horizontal gene transfer and evolutionary dynamics within microbiomes.

Methodology:

A random forest classifier was trained on 23,211 bacterial chromosome, plasmid, and bacteriophage sequences that were fragmented into 5,000-nucleotide segments and subdivided into k-mers to model k-mer distributions for classification.