Skip to main content
Fig. 6 | BMC Bioinformatics

Fig. 6

From: Magic-BLAST, an accurate RNA-seq aligner for long and short reads

Fig. 6

Performance of the aligners and quality control characteristics measured on the Baruzzo Illumina-like benchmark. First, the results of the comparison of all alignments to the truth are shown: a alignment precision, b alignment recall, counting as true an alignment which is identical or included in the truth (perfect or partial); each alignment for multiply aligned reads is counted. Second, the statistics of the alignment independently of the truth are shown: c percentage of reads aligned. d percentage of bases aligned. Note that the aligners differ on their view of what can be reported as an ‘alignment’ and on their thresholds for minimal aligned length and maximal level of mismatches. e Percentage of reads aligned uniquely. A higher rate of unique alignments is desirable, but the true number of ambiguous reads is unknown. f The number of mismatches per kilobase aligned. The true rate (first line of each dataset) was measured using the official benchmark mapping. A rate below the truth, correlated with lower alignment recall, characterizes an aligner which cannot deal with high levels of mismatches. g the intron discovery F-score, measured for all introns (coverage at least 1). Magic-BLAST reaches the highest F-score for intron discovery in all datasets

Back to article page