On the other hand, the primer check step for removing reads lacking both primer sequences had the possibility to incorrectly remove reads containing V1 V2 regions longer than the maximum length of 431 bp in the filter-passed reads.
The mapped reads were normalized between samples by uniformly removing reads to obtain an identical number of mapped reads for each experiment.
Because the samples were oversequenced (>3 million reads total), we removed reads that increased coverage over a 30× threshold, leaving 2 million reads for the assembly.
The workflow for obtaining 'clean' reads involved filtering low quality tags; removing raw reads with 5' primer contaminants; trimming 3' adaptors; removing reads without insert tags; discarding reads with polyA tails; and removing contaminants formed by adaptor adaptor ligation.
MarkDuplicates from Picard-tools [ 38] was used for removing read duplicates.
Adaptor sequences were removed, read quality was checked, and reads were trimmed and filtered for length.
Single reads were mapped using BWA (Version: 0.6.1-r104) (Li and Durbin, 2010) to the sacCer3 reference genome and were processed with samtools (Li et al., 2009) to remove duplicated reads for parallel analysis.
It oriented (for 16S rRNA) or translated (for protein families) the remaining reads by comparing them with BLAST against the simulated reference database, and removed any reads for which this could not be done accurately.
After removing rRNA reads, for each of the substrates tested, 74.3% to 84.2% of the reads were mapped to previously annotated coding regions, and the remaining were either upstream of a coding sequence (CDS; thus putatively identifying a 5′-untranslated region (5′-UTR)) or mapped to unannotated or potentially mis-annotated regions.
We used the pipeline proposed in [ 35] to remove low quality reads for de novo assembly.
