Table 1.

Base callers accuracy and runtime comparison

	Bustard		AYB		BlindCall slow		BlindCall fast		freeIbis
Perfect reads	1 446 079		1 532 000		1 509 451		1 508 779		1 530 099
Error rate (%)	0.29		0.21		0.23		0.23		0.21
Time (minimum)	17		217		8/12		4/8		9/126
Assembly results	N50	Maximum	N50	Maximum	N50	Maximum	N50	Maximum	N50	Maximum
5×	610	1122	628	1155	629	1164	623	1167	649	1184
10×	3 375	3469	3198	3322	3382	3487	3389	3485	3306	3418
20×	4466	4478	4627	4637	4511	4523	4470	4483	4333	4357

AYB, accuracy and run times for Bustard. freeIbis and BlindCall for a dataset of 1.9 million reads from a HiSeq 2000 run of PhiX174. BlindCall Fast corresponds to non-iterative version of the blind-deconvolution method. Running times for BlindCall are reported as (processing time/total time), where the total time includes reading intensity data from disk and writing base-calls to disk. For freeIbis, we report the time as (predicting time with single thread/ training time with 10 threads). BlindCall was able to produce base-calls of comparable accuracy to AYB and freeIbis at significantly faster computational time (8 min/12 min versus 217 min and 126 min, respectively). It is also faster than Bustard (8 min/12 min versus 17 min). AYB, freeIbis and BlindCall all improve on Bustard base calls. We also compared assemblies of the PhiX174 genome using reads generated by Bustard, BlindCall, freeIbis and AYB. The reported N50s and Max contig lengths are averages >100 random samples with the corresponding coverage (5×, 10× or 20×). While BlindCall is able to process data at a significantly lower computational cost, the assemblies obtained using BlindCall are of comparable quality to those obtained using AYB or freeIbis.