TOPAZ: asymmetric suffix array neighbourhood search for massive protein databases

BMC Bioinformatics

Table 1 Runtimes using different numbers of threads and overall sensitivity compared to BLAST results for all methods tested

		Runtime using N threads (seconds)							Mem.
Method	Sens.	1	2	4	8	16	32	64	(GB)
BLAST	1.0	1,801,293	948,124	479,400	247,715	128,401	70,107	43,647	9.9
DIAMOND (default)	0.840	41,863	21,330	10,677	5520	3269	2259	1919	8.6
DIAMOND (sensitive)	0.926	137,100	67,858	35,615	19,077	10,452	8463	7723	10.0
DIAMOND (more sens.)	0.931	166,737	85,689	45,954	23,716	12,849	11,204	10,174	12.0
Lambda (fast)	0.681	3325	1916	1152	776	586	443	435	104.2
Lambda (sensitive)	0.726	5200	2819	1636	940	631	477	456	104.3
LAST (m=100)	0.636	1714	946	525	326	233	166	141	20.9
LAST (m=1000)	0.838	7993	4397	2308	1381	859	585	439	24.3
LAST (m=10000)	0.881	33,411	17,859	9473	5224	3096	1891	1347	30.1
SANSparallel (fast)	0.696	6301	2938	1573	992	577	433	472	230.4
SANSparallel (slow)	0.758	11,258	5123	2801	1770	1150	985	1247	230.4
SANSparallel (verislow)	0.779	18,039	8396	4330	2639	1639	1305	1551	230.4
TOPAZ (fast)	0.800	2267	1111	569	299	160	99	78	57.2
TOPAZ (default)	0.840	5961	2961	1559	779	418	243	175	60.5

TOPAZ (default) has similar sensitivity to LAST (m=1000) and DIAMOND (default), but is faster than both methods irrespective of the number of threads. Bold indicates the fastest method for each number of threads. TOPAZ (fast) is the fastest method for 8–64 threads. LAST (m=100) is the fastest method for 1–4 threads, but suffers from the lowest sensitivity

ISSN: 1471-2105