Plasmidity: A Novel Pipeline for Plasmid Sequence Prediction

We have built PLASMIDITY, an automated pipeline to assemble short reads and characterize contigs searching for biologically relevant features for plasmid sequences detection applying machine learning methods, including contig length, multiplicity, circularity, insertion sequence composition, chromosome markers (rRNA, tRNA), plasmid markers (OriV, Rep, Mob/Rlx), G+C% variation, similarity with plasmid references, and k-mer profile. Using simulated WGS reads from 200 taxonomically diverse samples, we trained two Gradient Boosting classifiers, achieving a maximum F1 score of 0.901 in the test dataset. The comparison of the models against other plasmid prediction tools shows an improvement in the outcomes.