Dataset Information

Beating Naive Bayes at Taxonomic Classification of 16S rRNA Gene Sequences.

ABSTRACT: Naive Bayes classifiers (NBC) have dominated the field of taxonomic classification of amplicon sequences for over a decade. Apart from having runtime requirements that allow them to be trained and used on modest laptops, they have persistently provided class-topping classification accuracy. In this work we compare NBC with random forest classifiers, neural network classifiers, and a perfect classifier that can only fail when different species have identical sequences, and find that in some practical scenarios there is little scope for improving on NBC for taxonomic classification of 16S rRNA gene sequences. Further improvements in taxonomy classification are unlikely to come from novel algorithms alone, and will need to leverage other technological innovations, such as ecological frequency information.

SUBMITTER: Ziemski M

PROVIDER: S-EPMC8249850 | biostudies-literature | 2021

REPOSITORIES: biostudies-literature

ACCESS DATA

Publications

Beating Naive Bayes at Taxonomic Classification of 16S rRNA Gene Sequences.

Ziemski Michal M Wisanwanichthan Treepop T Bokulich Nicholas A NA Kaehler Benjamin D BD

Frontiers in microbiology 20210618

Naive Bayes classifiers (NBC) have dominated the field of taxonomic classification of amplicon sequences for over a decade. Apart from having runtime requirements that allow them to be trained and used on modest laptops, they have persistently provided class-topping classification accuracy. In this work we compare NBC with random forest classifiers, neural network classifiers, and a perfect classifier that can only fail when different species have identical sequences, and find that in some pract ...[more]

PMID: 34220738

Similar Datasets

Project description:Background. 16S rRNA gene sequences are routinely assigned to operational taxonomic units (OTUs) that are then used to analyze complex microbial communities. A number of methods have been employed to carry out the assignment of 16S rRNA gene sequences to OTUs leading to confusion over which method is optimal. A recent study suggested that a clustering method should be selected based on its ability to generate stable OTU assignments that do not change as additional sequences are added to the dataset. In contrast, we contend that the quality of the OTU assignments, the ability of the method to properly represent the distances between the sequences, is more important. Methods. Our analysis implemented six de novo clustering algorithms including the single linkage, complete linkage, average linkage, abundance-based greedy clustering, distance-based greedy clustering, and Swarm and the open and closed-reference methods. Using two previously published datasets we used the Matthew's Correlation Coefficient (MCC) to assess the stability and quality of OTU assignments. Results. The stability of OTU assignments did not reflect the quality of the assignments. Depending on the dataset being analyzed, the average linkage and the distance and abundance-based greedy clustering methods generated OTUs that were more likely to represent the actual distances between sequences than the open and closed-reference methods. We also demonstrated that for the greedy algorithms VSEARCH produced assignments that were comparable to those produced by USEARCH making VSEARCH a viable free and open source alternative to USEARCH. Further interrogation of the reference-based methods indicated that when USEARCH or VSEARCH were used to identify the closest reference, the OTU assignments were sensitive to the order of the reference sequences because the reference sequences can be identical over the region being considered. More troubling was the observation that while both USEARCH and VSEARCH have a high level of sensitivity to detect reference sequences, the specificity of those matches was poor relative to the true best match. Discussion. Our analysis calls into question the quality and stability of OTU assignments generated by the open and closed-reference methods as implemented in current version of QIIME. This study demonstrates that de novo methods are the optimal method of assigning sequences into OTUs and that the quality of these assignments needs to be assessed for multiple methods to identify the optimal clustering method for a particular dataset.

Project description:BackgroundThe phylogeny of the genus Methanobrevibacter was established almost 25 years ago on the basis of the similarities of the 16S rRNA oligonucleotide catalogs. Since then, many 16S rRNA gene sequences of newly isolated strains or clones representing the genus Methanobrevibacter have been deposited. We tried to reorganize the 16S rRNA gene sequences of this genus and revise the taxonomic affiliation of the isolates and clones representing the genus Methanobrevibacter.ResultsThe phylogenetic analysis of the genus based on 786 bp aligned region from fifty-four representative sequences of the 120 available sequences for the genus revealed seven multi-member groups namely, Ruminantium, Smithii, Woesei, Curvatus, Arboriphilicus, Filiformis, and the Termite gut symbionts along with three separate lineages represented by Mbr. wolinii, Mbr. acididurans, and termite gut flagellate symbiont LHD12. The cophenetic correlation coefficient, a test for the ultrametric properties of the 16S rRNA gene sequences used for the tree was found to be 0.913 indicating the high degree of goodness of fit of the tree topology. A significant relationship was found between the 16S rRNA sequence similarity (S) and the extent of DNA hybridization (D) for the genus with the correlation coefficient (r) for logD and logS, and for [ln(-lnD) and ln(-lnS)] being 0.73 and 0.796 respectively. Our analysis revealed that for this genus, when S = 0.984, D would be <70% at least 99% of the times, and with 70% D as the species "cutoff", any 16S rRNA gene sequence showing <98% sequence similarity can be considered as a separate species. In addition, we deduced group specific signature positions that have remained conserved in evolution of the genus.ConclusionsA very significant relationship between D and S was found to exist for the genus Methanobrevibacter, implying that it is possible to predict D from S with a known precision for the genus. We propose to include the termite gut flagellate symbiont LHD12, the methanogenic endosymbionts of the ciliate Nyctotherus ovalis, and rat feces isolate RT reported earlier, as separate species of the genus Methanobrevibacter.

Dataset Information

Beating Naive Bayes at Taxonomic Classification of 16S rRNA Gene Sequences.

Publications

Beating Naive Bayes at Taxonomic Classification of 16S rRNA Gene Sequences.

Similar Datasets

OmicsDI is part of the ELIXIR infrastructure

Tweets