Characterization of statistical features for plant microRNA prediction

BMC Genomics

Table 1 List of miRDeep parameters estimated for plant specific miRNAs.

Feature		Original parameters of miRDeep	Plant specific (monocot)	Dicot specific (if any)
MFE	Known	Cumulative Distribution Function: $F (x) = e^{- e^{(x - l o c a t i o n) / s c a l e}}$ Location = 32; Scale = 5.5	Log-odds score: $f (x) = \frac{a}{(b + e^{x * c})}$	Log-odds score: $f (x) = \frac{a}{(b + e^{x * c})}$
	Background	Cumulative Distribution Function: $F (x) = e^{- e^{(x - l o c a t i o n) / s c a l e}}$ Location = 23; Scale = 4.8	a = 1.339e-12 b = 2.778e-13 c = 45.843	a = 4.46e-4 b = 9.125e-5 c = 26.929
Stability (log-odds)	Stable	1.6	1.37	0.63
	Unstable	-2.2	-3.624	-3.17
Nucleus conservation (log-odds)	Conserved	3	7.63
	Non-conserved	-0.6	-1.17
Excision length		140 nt	300 nt
Paired	Total	≥14	≥15 nt
Unpaired	Total	NA	≤5 nt
	Consecutive unpaired	NA	≤3 nt
Bulge	Total	could be as high as 5 nt ^#	≤2 nt
Maximum multiple hits of deep-seq read		5	20

The cumulative distribution of MFE of known and background precursors, in miRDeep algorithm, are represented by Gumbel distribution functions, where F(x) is the cumulative frequency of MFE less than or equal to variable x, and location is used for mean centering the distribution of MFE. On the other hand, in the parameterized algorithm, log-odds of MFE were directly obtained from the distributions of real and background precursors, and then represented by a modified sigmoid function where f(x) denotes log-odds of a given normalized MFE. ^#The size of bulge was estimated based on allowed difference between length of mature and star sequences in miRDeep. The bulges often appear when one of the duplex sequences has additional bases than its counterpart.

ISSN: 1471-2164