The effects of contig length and depth on the estimation of SNP frequencies, and the relative abundance of SNPs in protein-coding and non-coding transcripts of tiger salamanders (Ambystoma tigrinum)

Table 2 Model selection among candidate models predicting frequencies of SNPs, transition and transversion in contigs

Model		Parameters^a	AIC	Δ AIC^b	w_i^c
for estimating the number of SNPs
	M1 (best, full model)	Intercept*, LENGTH, DEPTH**, C/NC†	7283.4	0.0	0.904
	M2	Intercept, DEPTH, C/NC	7287.9	4.5	0.096
	M3	Intercept, LENGTH, C/NC	7540.7	257.3	1.2 × 10^-56
	M4	Intercept, C/NC	7748.3	464.9	1.0 × 10^-101
for estimating the number of transitions (T _i )
	M1 (best, full model)	Intercept, LENGTH†, DEPTH, C/NC	5682.4	0.0	0.681
	M2	Intercept, DEPTH, C/NC	5683.9	1.5	0.319
	M3	Intercept, LENGTH, C/NC†	5896.1	213.7	2.7 × 10^-47
	M4	Intercept, C/NC	6061.0	378.5	6.3 × 10^-83
for estimating the number of transversions (T _v )
	M1 (best, full model)	Intercept, LENGTH, DEPTH*, C/NC	4077.1	0.0	0.991
	M2	Intercept, DEPTH, C/NC†	4086.6	9.5	0.009
	M3	Intercept, LENGTH, C/NC	4183.8	106.7	6.7 × 10^-24
	M4	Intercept*, C/NC	4310.4	233.3	2.2 × 10^-51

Model selection among candidate regression models using negative binomial distribution predicting the frequency of SNPs, transitions, and transversions in contigs (depth of 10 reads or more; length of 501 bp or longer), using Akaike's Information Criteria (AIC).
^a LENGTH and DEPTH are the length and the depth of contigs, respectively, and C/NC is a dummy variable for the type of transcript (protein-coding (coded with 1) vs non-coding transcript (coded with 0). **, P < 0.01; *, P < 0.05; †, P < 0.1) variable in each model.
^bΔ AIC is the difference between the AIC of the best fitting model and that of each model.
^cw_i is Akaike weight of each model.

ISSN: 1471-2164