tRNA read alignment with Bowtie and Bowtie 2

To test previously used alignment strategies as in DM-tRNAseq (Zheng et al., 2015) or ARM-seq (Cozen et al., 2015), a non-redundant set of reference human tRNA transcripts was created by fetching the full set of 610 predicted tRNA genes for human genome hg19 from GtRNAdb (Chan and Lowe, 2016) and the 22 mitochondrially encoded human tRNA genes from mitotRNAdb (Jühling et al., 2009). Following intron removal and addition of 3′ CCA (for nuclear-encoded transcripts) and 5′-G (for tRNA-His), a curated set of 596 genes (excluding anticodon/isotype mismatch and nuclear-encoded mitochondrial tRNAs) were collapsed into 420 unique sequences. Corresponding Bowtie and Bowtie 2 indices were built from this set of references. Bowtie alignment was performed with a maximum of 3 allowed mismatches per read (-v 3), filtering for uniquely aligning reads (-m 1) and ensuring the best alignment from the best stratum (i.e., reads with the least number of mismatches) were reported (--best --strata). Bowtie 2 alignments were performed in very sensitive local mode (--very-sensitive --local) and up to 100 alignments per read were allowed (-k 100). Read quality scores were ignored for alignment score and mismatch penalty calculation (--ignore-quals) with increased penalties for ambiguous characters (“N”) in reference or read (--np 5). Output alignments in SAM format were reordered to match read order in input fastq file (--reorder). The alignment commands for both algorithms are given below:

bowtie -v 3 -m 1 --best --strata --threads 40 -S

bowtie2 --local -x -k 100 --very-sensitive --ignore-quals --np 5 --reorder -p 40 -U

QuantM-tRNAseq data for HEK293 T-Rex Flp-IN cells downloaded from the NCBI Gene Expression Omnibus repository was adapter-trimmed and analyzed with Bowtie 2 as described in (Pinkard et al., 2020):

bowtie2 --local --score-min G,1,8 -D 20 -R 3 -N 1 -L 10 -i S,1,0.5