Plasmid Host Attribution with ML

One-Sentence Definition

Plasmid host attribution with machine learning predicts which bacterial host taxon or lineage is the most likely carrier of a plasmid sequence — critical when metagenomes or fragmented assemblies disconnect plasmids from chromosomes.

Simple Explanation

You found a resistance plasmid in the data soup; ML guesses which bacterium it probably came from.

Detailed Scientific Explanation

Features used:

Feature classExamples
Sequence compositionk-mer, GC, codon usage
Replicon / MOB typeInc groups, relaxases
Gene contentAMR/virulence cargo
Graph linkageHi-C, proximity ligation (experimental)
Host CRISPR spacersMatching plasmid motifs

Models range from random forests on composition to deep classifiers on contig embeddings (DNA and Genome Language Models). Evaluation needs careful separation of plasmid families to avoid leakage.

Mechanism

Represent plasmid contig → classify host taxon/lineage → calibrate confidence → report with uncertainty for epi use.

Clinical Importance

  • Metagenomic resistome tracking in hospitals/wastewater (Metagenomics, One Health).
  • Supports infection control when culture fails but plasmid signatures appear.

Research Importance

  • Plasmid ecology and host range evolution.

Diagnostic Relevance

  • Emerging for mNGS interpretation; confidence thresholds essential.

AMR Relevance

High for connecting mobile bla/mcr genes to likely bacterial reservoirs.

Active Recall Questions

  1. Why is host attribution hard in short-read metagenomes?
  2. Name two feature classes for plasmid–host models.
  3. How do CRISPR spacers help attribution?

Connections