To compare the goodness-of-fit of models based on different distributions, we used Vuong’s non-nested hypothesis test [21]. BIC-corrected Vuong z-statistic was used to impose stronger penalty on additional parameters. The pairwise comparison results are summarized in Table 1. Poisson distribution GLM is treated as the reference distribution to compare to, given its simple configuration. As expected, NB GLM is superior to Poisson GLM, since NB models dispersion of the data, and this is also supported by dispersion test [22] (z = 68.5881, p value < 2.2e-16). The necessity of modeling zero-inflation is supported by the Vuong’s test comparing ZIP with Poisson GLM. When comparing ZIP with NB, NB fits the data better. However, it is worth noting the evidence of superiority – the absolute value of BIC-corrected Vuong z-statistic – is much smaller than the other tests. The merit of considering both dispersion and zero-inflation is further emphasized by the comparisons of ZIP with ZINB and NB with ZINB. In conclusion, based on Vuong’s test, for Ion Proton sequencing dataset, the most appropriate distribution is ZINB, followed by NB, ZIP
