NoDe: a fast error-correction algorithm for pyrosequencing amplicon reads

BMC Bioinformatics

Table 2 Tabular overview of the computational cost of the different denoising algorithms

Denoising approach	SFF extraction	Trim reads	Denoising algorithm (average memory)	Aligning	Filter alignment	Total time
NoDe	00:00:12	00:00:02	00:02:16 (760 MBs)	00:06:40	00:01:00	00:10:05
Pre-cluster	00:00:12	00:00:02	00:00:13 (100 MBs)	00:06:40	00:01:00	00:08:02
AmpliconNoise	00:00:12	00:00:01	08:25:17 (1,900 MBs)	00:03:40	00:01:00	08:30:27
Denoiser	00:00:12	00:00:01	00:38:17 (2,300 MBs)	00:02:30	00:01:00	00:42:00
Acacia	00:00:12	00:00:02	00:00:55 (1,600 MBs)	00:06:40	00:01:00	00:08:49

To have an idea about the computational cost for each step, the complete pipeline was subdivided in different steps to illustrate its running time, as described above. From the table, it can be observed that the computational burden added to the complete preprocessing pipeline (by integrating the NoDe algorithm) was relatively small, and it was largely compensated with a significant improvement in the error rate, that exceeded the second best performing (but computationally intensive) algorithm AmpliconNoise. For the denoising algorithms, the average amount of memory required was added.

ISSN: 1471-2105