Statistical modeling of STR capillary electrophoresis signal

BMC Bioinformatics

Table 1 Models for Peaks and Drop-out components

Component	Model	Input	Likelihood function
True peak	\(\mathcal {N}(\mu,\sigma) ; \left \{\begin {array}{l} \mu =u(x) = a.x +b \\ \sigma =v(x) = c.x +d \end {array}\right.\)	\({x_{i}=A_{c}.e^{B_{c}.s_{i}}}\phantom {\dot {i}\!}\)	\(\mathcal {L}_{h}\)
Noise peak
Forward stutter		x_i=PPH_i
Reverse stutter
True peak D.O.	p(x)=a.e^b.x	\(\phantom {\dot {i}\!}{x_{i}=A_{c}.e^{B_{c}.s_{i}}}\)	\(\mathcal {L}_{do}\)
Reverse stutter D.O.		x_i=PPH_i
Forward stutter D.O.
Noise peak D.O.	p(x)=a	a=f(h_i)

For each model component, at each locus, we indicate the probability distribution, its analytical form, and the model input x_i, namely Decayed Amplitude for peaks in allelic and noise position, and PPH_i (parent peak height) for peaks in reverse and forward stutter position. Peak models follow a normal density, and the frequencies of drop-out are modeled using an exponential decay. Noise drop-out parameter a is independent of the observed sample. D.O. denotes drop-out

ISSN: 1471-2105