Identifying targets of multiple co-regulating transcription factors from expression time-series by Bayesian model comparison

Table 2 Full model ChIP evaluation bootstrap results

Validating both positive and negative predictions
Top 200					Top 400					Top 800
	MAP	ML	Reg	Inf		MAP	ML	Reg	Inf		MAP	ML	Reg	Inf
MAP		***	**	-	MAP		***	***	-	MAP		***	***	-
ML				-	ML				-	ML				-
Reg		+		-	Reg				-	Reg				-
Inf	-	-	-	-	Inf	-	-	-	-	Inf	-	-	-	-
Top 1600					Top 3200					Top 6003
	MAP	ML	Reg	Inf		MAP	ML	Reg	Inf		MAP	ML	Reg	Inf
MAP		***	***	-	MAP		***		-	MAP		***		***
ML				-	ML				-	ML
Reg		*		-	Reg	***	***		-	Reg	***	***		***
Inf	-	-	-	-	Inf	-	-	-	-	Inf		*
Validating only positive predictions
Top 200					Top 400					Top 800
	MAP	ML	Reg	Inf		MAP	ML	Reg	Inf		MAP	ML	Reg	Inf
MAP		.	***	-	MAP		**	***	-	MAP		***	***	-
ML			**	-	ML			***	-	ML			***	-
Reg				-	Reg				-	Reg				-
Inf	-	-	-	-	Inf	-	-	-	-	Inf	-	-	-	-
Top 1600					Top 3200					Top 6003
	MAP	ML	Reg	Inf		MAP	ML	Reg	Inf		MAP	ML	Reg	Inf
MAP		***	***	-	MAP		***	***	-	MAP		***	*	***
ML			***	-	ML			***	-	ML				***
Reg				-	Reg				-	Reg		***		***
Inf	-	-	-	-	Inf	-	-	-	-	Inf

The results of 100,000-fold bootstrap resampling of the data set of observed genes to assess significance of differences in ranking method performance. The methods studied are: MAP = MAP GP method, ML = ML Baseline method, Reg = regression baseline method, Inf = Inferelator. For each pair of methods, the marks in the tables show how often the method on the corresponding row dominated the one on the corresponding column. The marks are interpreted as follows: ‘.’: >80% dominance, ‘+’: >90% dominance, ‘*’: >95% dominance, ‘**’: >99% dominance, ‘***’: >99.9% dominance, ‘-’: comparison not applicable.

ISSN: 1752-0509