EufindtRNA was implemented from the published algorithm by Pavesi and colleagues [Pavesi et al., 1994]. The step-wise algorithm uses four probabilistic profiles for identifying basic tRNA features: `A box' nucleotide composition, `B box' composition, nucleotide distance between identified A and B boxes, and distance between identified B boxes and RNA polymerase III termination signals (four or more consecutive thymine nucleotides). In a search, an ``intermediate'' score is obtained by adding scores from identified A and B boxes to the score for the nucleotide distance between them. A final score is obtained by adding the intermediate score to the score for the distance to the nearest termination signal. If the final score is above a specific cutoff, the tRNA identity and location are saved.
Scores from over 30 example tRNAs described in the original publication match our implementation to within 0.1 log odds units. tRNAscan-SE uses a less selective version of the algorithm described above which does not search for transcription termination signals; instead, the intermediate score is used as a final cutoff. Also, the intermediate score cutoff is loosened slightly to -32.10 relative to the intermediate cutoff described in the original algorithm, -31.25. Although the program is designed for eukaryotic tRNA detection, we found EufindtRNA to be effective at identifying prokaryotic tRNAs if the intermediate cutoff score is further adjusted. tRNAscan-SE has a specific option (-P) for scanning prokaryotic sequences which loosens the intermediate cutoff score to -36.0. Also, as with tRNAscan 1.4, ambiguous nucleotides are automatically assigned the best of the four non-ambiguous nucleotide scores at that position in the scoring matrices.