LtrDetector: A tool-suite for detecting long terminal repeat retrotransposons de-novo

BMC Genomics

Table 4 Results on six plant genomes: We tested three tools on one model organism, A. thaliana, and five important crops of varying genomic size and repeat content

Tool	Total	TP	GT	FP	Sensitivity	Precision	F1	Time (hr:min:sec)	Memory (GB)
A. thaliana
LTR_Finder	399	106	248	0	0.427	1.000	0.599	0:30:46	0.86
LTRharvest	2301	180	248	6	0.726	0.968	0.829	0:01:08	0.24
LtrDetector	1714	187	248	9	0.754	0.954	0.842	0:04:02	4.45
O. sativa
LTR_Finder	5324	1163	1760	14	0.661	0.988	0.792	5:19:03	0.95
LTRharvest	9761	1392	1760	182	0.791	0.884	0.835	0:03:09	0.34
LtrDetector	7343	1442	1760	119	0.819	0.924	0.868	0:15:30	5.31
S. bicolor
LTR_Finder	11734	4219	6565	67	0.643	0.984	0.778	10:43:26	1.62
LTRharvest	22700	4476	6565	502	0.682	0.899	0.776	0:04:23	0.60
LtrDetector	24682	5285	6565	214	0.805	0.961	0.876	1:10:47	6.1
G. max
LTR_Finder	12141	1748	3130	7	0.558	0.996	0.716	25:18:21	1.88
LTRharvest	29016	2171	3130	20	0.694	0.991	0.816	0:09:35	0.48
LtrDetector	25537	2542	3130	12	0.812	0.995	0.894	0:43:06	6.11
Z. mays
LTR_Finder	60860	11411	16839	13	0.678	0.999	0.807	111:21:37	12.36
LTRharvest	101943	11244	16839	102	0.668	0.991	0.798	0:15:00	2.36
LtrDetector	116923	13122	16839	71	0.779	0.995	0.874	5:53:08	9.62
H. vulgare
LTR_Finder	–	–	–	–	–	–	–	–	–
LTRharvest	207016	4378	9164	492	0.478	0.899	0.624	1:33:29 *	5.12
LtrDetector	213367	6824	9164	199	0.745	0.972	0.843	17:24:04**	14.15
Total (Excluding H. vulgare)
LTR_Finder	90458	18647	28542	101	0.653	0.995	0.789	153:13:13	–
LTRharvest	165721	19463	28542	812	0.682	0.960	0.797	0:33:15	–
LtrDetector	176197	22578	28542	425	0.791	0.982	0.876	8:06:33	–
Total (Including H. vulgare)
LTRharvest	372737	23841	37706	1304	0.632	0.948	0.759	02:06:44	–
LtrDetector	389564	29402	37706	624	0.780	0.979	0.868	25:30:37	–

Parameters used for each tool can be found in the “Implementation” section. We used an additional utility to process each of LTR_Finder and LTRharvest in parallel because neither supports multi-threading. We did so to ensure fair comparison in terms of time since our tool, LtrDetector, is concurrent by default. Total is the number of proposed LTR-RTs, TP is number of true positives, GT is number of elements in the ground truth, FP are false positives. Sensitivity, Precision, and F1 are defined by Eqs. 1, 2, and 3. We report all measures for each genome and in total. Note: Results for LTR_Finder are unavailable for the Hordeum vulgare (barley) genome because memory demands repeatedly caused the computer to crash on four computer cores, and a subsequent trial on one core was unable to finish over two weeks of run time (2/7 chromosomes finished). All trials run on four cores unless otherwise noted. * LTRharvest run on one thread for H. vulgare. ** LtrDetector run on three threads for H. vulgare

ISSN: 1471-2164