In silico miRNA prediction in metazoan genomes: balancing between sensitivity and specificity

BMC Genomics

Table 2 Scoring model performance

Scoring model description	# ^a	Wt ^b	LDF ^c	Chi – Square ^d	AUC^e	AUC ^e	AUC ^e	Selectivity ^f
					Nematoda	H. sapiens	Metazoa	95%	75%
Default models ^g
Metazoa	3,902	-	0.95	9.95e-7	0.9760	0.9764	0.9814	12.57	74.06
H. sapiens	781	-	0.95	0.090	0.9733	0.9794	0.9806	11.04	68.90
C. elegans	131	-	0.95	0.405	0.9747	0.9638	0.9735	7.73	41.43
Optimized models ^g
Metazoa	3902	Y	0.90	9.95e-7	0.9813	0.9848	0.9874	21.92	105.7
H. sapiens	781	Y	0.90	0.090	0.9798	0.9870	0.9871	21.93	87.36
C. elegans	131	Y	0.90	0.405	0.9817	0.9775	0.9835	16.65	75.34

^a Number of miRNAs in taxonomic set
^b -; No weighting (weight = 1), Y; weighting individual descriptors by the square root of their discriminative power (Table 1)
^c LDF parameterized at 95% or 90% of the CDF
^d Goodness-of-fit is evaluated by averaging Chi-square test statistics of all 40 descriptors
^e A rea u nder the c urve of ROC curves measured on the taxonomic sets of known miRNA hairpins from Nematodes (211 miRNA hairpins), H. sapiens (781) and Metazoa (3,902) versus 200,000 randomly selected genomic hairpins from C. elegans.
^f Selectivity expressed at 95% and 75% sensitivity, with sensitivity measured on the taxonomic set of Metazoa, specificity measured on set of 3,526,115 C. elegans genomic hairpins.
^g Scorings models composed of the subset of 18 most informative descriptors

ISSN: 1471-2164