Table 3.

Percentage of correctly identified sites for different values of λ and length L

		L
crp	λ	400	500	600	700
	0	47	0	0	0
	10	53	47	0	0
	20	53	47	0	0
	30	53	47	47	0
	50	0	0	0	0
	100	0	0	6*	6*
	200	0	0	0	0
	500	0	0	0	0
		L
abf1	λ	600	700	800	900
	0	56	0	0	0
	10	56	0	0	0
	20	56	56	50	0
	30	56	56	50	0
	50	50	0	0	0
	100	0	0	0	0
	200	0	0	0	0
	500	0	0	0	0
		L
rap1	λ	1,400	1,500	1,600	1,700
	0	80	0	0	0
	10	80	0	0	0
	20	80	0	0	0
	30	73	80	67	0
	50	73	80	67	0
	100	73	0	67	0
	200	73	73	67	0
	500	67	60	67	0
		L
reb1	λ	500	600	700
	0	86	0	0
	10	86	64	0
	20	79	64	0
	30	0	0	0
	50	0	7*	0
	100	0	7*	0
	200	0	0	0
	500	0	0	0

Rows show the results with the specified λ value and columns correspond to different values of L. The variable change point model and normal prior is used with motif width and motif specification (uni- or bimodal) listed in Table 1. For each dataset L, the top 2L starting points were used according to the MEME starting point selection procedure. For the rap1 test set, a different set of random starting points was used for each L. Entries labeled with an asterisk correspond to a trial in which the correct motif was not found, but one or two sites were correctly predicted by chance, with a spurious motif.