OTU clustering

The OTU clustering tool clusters a collection of reads to operational taxonomic units.

To run the tool, go to

        Toolbox | Microbial Genomics Module (Image mgm_folder_closed_flat_16_h_p) | Metagenomics (Image wma_folder_open_flat_16_n_p) | Amplicon-Based Analysis (Image otutools_open_16_n_p) | OTU clustering (Image OTU_clustering_16_n_p)

The tool aligns the reads to reference OTU sequences (e.g. the reference database) to create an "alignment score" for each OTU. If the input sequence is shorter, the unaligned ends of the reference are ignored. For example, if a shorter sequence has 100% identity to a fragment of a longer reference sequence, the tool will assign 100% identity and assign the read to the OTU. In the opposite case (longer read mapping to short database reference), the unaligned ends will count as indels, and the percentage identity will be lower.

When the input consists of paired reads, the OTU clustering tool will initially group them into pairs, and align both reads of a pair to the same OTUs. Both reads of a pair will be assigned to the one OTU where they BOTH align with the highest identity possible. Finally, the tool merges both reads of the pair using a stretch of N to the fragments so that the paired read looks as much as possible like the OTU they have been assigned to. For example, the forward-reverse pair (ACGACGACG, GTAGTAGTA) will be turned into ACGACGACGnnnnnnnnnnnnnnnnnnnnTACTACTAC. Reads that cannot be merged will be independently aligned to reference OTUs.

If a read due to insufficient similarity cannot be included in an already existing OTU, the algorithm attempts to optimize the alignment score by allowing "crossover" from one database reference to another at a cost (the chimera crossover cost). To speed up the chimera crossover detection algorithm, the read is not aligned to all OTUs but only to the most promising candidates found via a k-mer search. If the best match has at least one crossover and the "constructed alignment" meets the similarity percentage threshold, the read is considered chimeric.

By default, the similarity percentage parameter is set to 97% in the OTU Clustering tool. Therefore without the chimera crossover cost, the constructed alignments difference score can only be 3% at most. The smaller the chimeric cost, the more likely it is that a read is deemed chimeric; setting it too high decreases the chimeric detection.

To add samples to existing OTU clustering results, we recommend to run OTU clustering on the new samples separately and use the tool Merge Abundance Tables described in section Merge Abundance Tables to merge the OTU tables. If re-running analysis is necessary and you wish to compare with previous results, you should keep the original sample input order. Due to the iterative nature of the clustering algorithm, changing the order of input files can lead to slightly different results. In most conceivable cases that difference does not matter, specifically when using taxonomy informed and abundance weighted distance metrics like weighted UniFrac.