Vertebrate Mitochondrial tRNA Annotation

Mitochondrial tRNA genes (mt-tRNAs) in 1085 vertebrate species were obtained from mitotRNAdb (17 (link)) as the training set for building covariance models. In the mitochondria of vertebrates, there is typically one tRNA for each isotype, except mt-tRNA^Leu and mt-tRNA^Ser which each have two tRNAs with different anticodons (66 (link)). To achieve high specificity for each mt-tRNA type, we grouped the sequences by isotypes and anticodons resulting in 22 sets of mt-tRNA genes for generating individual covariance models. To further increase prediction accuracy for mammalian mt-tRNAs, we built a second set of covariance models that only utilized mt-tRNA genes from 282 mammals as the training set. Both sets of covariance models were built with two rounds of training. First, each subset of isotype-specific mt-tRNA sequences, excluding mt-tRNA^Ser(GCU), was aligned to the general covariance model from the original tRNAscan-SE (1 (link)) using Infernal (20 (link)). The general model was chosen because it was used in the organellar search mode of tRNAscan-SE v1.3 and has the greatest diversity of tRNA training sequences from all three domains of life. To accommodate the atypical secondary structure of mt-tRNA^Ser(GCU), a custom covariance model without the D-arm was built and used for aligning mt-tRNA^Ser(GCU) genes. The resulting alignments were used to build the first set of 22 isotype-specific covariance models for vertebrates and mammals. For the second round of training, the isotype-specific mt-tRNA sequences were aligned to the corresponding isotype-specific covariance models generated in the first round. These alignments were then used to build the final sets of mt-tRNA covariance models.