BACKGROUND: Plasmids and viruses are two types of mobile genetic elements (ME), that rely on host cells to reproduce and propagate themselves. Recently, metagenomics has greatly facilitated the discovery and characterization of new plasmids and viruses, which relies on accurate identification of these reads in metagenomes. Some state-of-the-art tools can identify plasmid or viral reads, while others are able to identify the probable host or source species of these reads. Since the Minimizer-based Naïve Bayes Classifier (MNBC) tool accurately classifies chromosomal and viral reads to the species level, we extended it to develop the MNBC-ME tool that can also identify plasmid reads and their putative host species. RESULTS: A standard reference- and test-sequence framework using simulated variable-length reads was used to benchmark MNBC-ME with eleven other state-of-the-art tools for ME identification: DeepMicroClass, geNomad, PPR-Meta, viralVerify, Plasmer, PlasClass, PlasX, VIBRANT, DeepVirFinder, HOTSPOT, and MOSTPLAS. MNBC-ME was the most consistent tool at classifying chromosomal, viral and plasmid reads of variable lengths, in contrast to the other tools whose precision or recall dropped below 50% in some circumstances. MNBC-ME also exceeded 65% and 70% performance in predicting host genus and family of plasmid reads, respectively. CONCLUSIONS: MNBC-ME is tool for identification of both short and long viral- and plasmid-originated reads across a wide variety of read types. It also identifies potential low-level host taxa for plasmid reads, and source taxa for chromosomal and viral reads. It is freely available at https://github.com/ComputationalPathogens/MNBC-ME and can be found as the 'mnbc-me' package in bioconda.

MOTIVATION: State-of-the-art tools for classifying metagenomic sequencing reads provide both rapid and accurate options, although the combination of both in a single tool is a constantly improving area of research. The machine learning-based Naïve Bayes Classifier (NBC) approach provides a theoretical basis for accurate classification of all reads in a sample. RESULTS: We developed the multithreaded Minimizer-based Naïve Bayes Classifier (MNBC) tool to improve the NBC approach by applying minimizers, as well as plurality voting for closely related classification scores. A standard reference- and test-sequence framework using simulated variable-length reads benchmarked MNBC with six other state-of-the-art tools: MetaMaps, Ganon, Kraken2, KrakenUniq, CLARK, and Centrifuge. We also applied MNBC to the "marine" and "strain-madness" short-read metagenomic datasets in the Critical Assessment of Metagenome Interpretation (CAMI) II challenge using a corresponding database from the time. MNBC efficiently identified reads from unknown microorganisms, and exhibited the highest species- and genus-level precision and recall on short reads, as well as the highest species-level precision on long reads. It also achieved the highest accuracy on the "strain-madness" dataset. AVAILABILITY AND IMPLEMENTATION: MNBC is freely available at: https://github.com/ComputationalPathogens/MNBC.

BACKGROUND: Identification of biomarkers, which are measurable characteristics of biological datasets, can be challenging. Although amplicon sequence variants (ASVs) can be considered potential biomarkers, identifying important ASVs in high-throughput sequencing datasets is challenging. Noise, algorithmic failures to account for specific distributional properties, and feature interactions can complicate the discovery of ASV biomarkers. In addition, these issues can impact the replicability of various models and elevate false-discovery rates. Contemporary machine learning approaches can be leveraged to address these issues. Ensembles of decision trees are particularly effective at classifying the types of data commonly generated in high-throughput sequencing (HTS) studies due to their robustness when the number of features in the training data is orders of magnitude larger than the number of samples. In addition, when combined with appropriate model introspection algorithms, machine learning algorithms can also be used to discover and select potential biomarkers. However, the construction of these models could introduce various biases which potentially obfuscate feature discovery. RESULTS: We developed a decision tree ensemble, LANDMark, which uses oblique and non-linear cuts at each node. In synthetic and toy tests LANDMark consistently ranked as the best classifier and often outperformed the Random Forest classifier. When trained on the full metabarcoding dataset obtained from Canada's Wood Buffalo National Park, LANDMark was able to create highly predictive models and achieved an overall balanced accuracy score of 0.96 ± 0.06. The use of recursive feature elimination did not impact LANDMark's generalization performance and, when trained on data from the BE amplicon, it was able to outperform the Linear Support Vector Machine, Logistic Regression models, and Stochastic Gradient Descent models (p ≤ 0.05). Finally, LANDMark distinguishes itself due to its ability to learn smoother non-linear decision boundaries. CONCLUSIONS: Our work introduces LANDMark, a meta-classifier which blends the characteristics of several machine learning models into a decision tree and ensemble learning framework. To our knowledge, this is the first study to apply this type of ensemble approach to amplicon sequencing data and we have shown that analyzing these datasets using LANDMark can produce highly predictive and consistent models.