Transcription factor target prediction using multiple short expression time series from Arabidopsis thaliana

BMC Bioinformatics

Table 1 Performance on real data

TF	Targetpool			Cor		ACT		CERMT-0		CERMT		CERMT Diagnostics
	Source	E	Size	Hits¹⁰⁰	P _Spec	Hits¹⁰⁰	P _Spec	Hits¹⁰⁰	P _Spec	Hits¹⁰⁰	P _Spec	Size	Hits^Gap	Gap
Induced targets
AREB	OX	0.16	28	2	0.03	1	0.15	9	0.03	9	0.04	128	10	0.20
CBF	CRE	14.32	2508	18	0.17	0	1.00	34	0.10	58	0.01	86	53	0.27
	OX	0.82	143	5	0.08	0	1.00	19	0.10	56	0.00		52
DREB2A	OX	0.12	21	3	0.03	1	0.07	10	0.06	11	0.05	244	15	0.04
HSFA2	OX	0.24	42	12	0.00	0	1.00	22	0.00	20	0.02	76	19	0.09
HY5	ChIP/KO	0.76	133	0	1.00	9	0.03	9	0.11	12	0.04	902	32	-0.19
	KO	0.69	120	13	0.00	19	0.00	14	0.00	14	0.00		46
	CRE	12.07	2113	22	0.11	24	0.05	19	0.31	19	0.30		153
MBF1c	OX	0.86	150	0	1.00	2	0.17	2	0.36	4	0.22	102	4	0.09
MYB2/MYC2	CRE	6.75	1182	2	0.46	6	0.64	5	0.39	4	0.42	536	31	0.23
	CRE	19.90	3485	31	0.02	23	0.32	30	0.07	32	0.05		121
	OX	0.15	26	0	1.00	0	1.00	1	0.26	1	0.30		2
NAC019	OX	0.08	14	0	1.00	0	1.00	0	1.00	1	0.13	33	0	0.01
NAC055	OX	0.05	9	0	1.00	2	0.00	0	1.00	0	1.00	121	0	0.09
NAC072	OX	0.13	23	0	1.00	0	1.00	2	0.09	5	0.04	96	5	0.25
PAP1	OX	0.24	42	1	0.08	8	0.00	0	1.00	0	1.00	27	0	0.28
ZAT12	OX	0.79	139	3	0.08	2	0.17	8	0.19	10	0.06	536	19	0.08
Repressed targets
CBF	OX	0.24	43	0	1.00	0	1.00	0	1.00	2	0.17	128	2	0.09
MBF1c	OX	0.46	80	0	1.00	0	1.00	0	1.00	1	0.47	1000	10	-0.19
ZAT12	OX	0.90	158	1	0.18	3	0.03	7	0.16	11	0.12	1000	33	0.01

Hit ratios based on real data for the top 100 genes associated with 12 different transcription factors (TFs). Bold entries are the highest values for that target pool, italic entries are insignificant over-representations according to Fisher's exact test (p ≥ 0.05). Targetpool size is the number of the true targets in our expression set, E is the expected number of hits when picking 100 genes genes at random and P _specis an empirical P-value indicating the probability that a random TF would give the same or better hit ratio. For CERMT, the best cluster size as indicated by the Gap statistic is shown along with the number of true targets for that regulon size. A large Gap statistic (greater than zero) indicates that the suggested regulon is significantly more related to the expression of the TF than could be expected from the expression of a shuffled TF. Target pools were defined from over-expression experiments (OX), knock-out experiments (KO), ChIP-chip experiments (ChIP) or by all genes carrying a known cis-regulatory element (CRE).

ISSN: 1471-2105