Table 1.
Alignment software undergoing complete evaluation.
Figure 1.
Workflow of alignment evaluation.
Reads were simulated from chromosome 21 of the ENSEMBL human transcriptome using dwgsim and aligned to the transcriptome with various parameter combinations for each alignment tool. The best parameter combination for each aligner was selected based on -measure first and runtime and memory consumption second in case of ties. The best parameters were then used to evaluate the alignment algorithms on larger test sets simulated from chromosomes 1-22, excluding chromosome 21.
Table 2.
Parameter Sensitivity.
Figure 2.
Influence of parameter selection.
Precision (x-axis) and recall (y-axis) are shown for SOAP2 (black triangles) and Bowtie 2 (gray diamonds) alignment tools at a read length of 72 bp. Different parameters had only little impact on the alignment performance of Bowtie 2 whereas the recall of SOAP2 was scattered widely. This is reflected by the high -measure dispersion value of SOAP2.
Table 3.
Evaluation on test set.
Figure 3.
Sensitivity of alignment algorithms to errors and indels.
The dependence of alignment precision (dotted lines ), recall (dashed lines
) and
-measure (solid lines –) on the number of sequencing errors (gray) and indels (black) was evaluated for each alignment algorithm analyzed in this study. In this figure, results are shown for the 72 bp long reads from the test set (5 million reads), excluding those algorithms which exceeded the memory and runtime cap. Alignment algorithms were sorted by algorithmic design: (A) Read indexing based aligners, (B) Reference indexing based aligners, (C) FM-Index based aligners.