MCMC implementation of the optimal Bayesian classifier for non-Gaussian models: model-based RNA-Seq classification

Table 1 Posterior predictive model diagnostics are given for 10 randomly selected genes from adenocarcinoma TCGA samples

Gene ID	IQR ( S _n)	95% int. for IQR ( x ^rep)	p-value
UPK1A\|11045	2.12	[1.0, 3.0]	0.09
OR4P4\|81300	0.00	[0.0, 0.0]	0.50
PCDHA12\|56137	139.22	[107.8, 187.0]	0.54
MDS2\|259283	1.85	[2.0, 5.0]	1.00
AXIN2\|8313	347.69	[331.5, 439.3]	0.85
DYNLT1\|6993	848.41	[830.0, 1043.3]	0.90
RARA\|5914	786.43	[706.8, 881.3]	0.62
TMEM194A\|23306	396.06	[367.0, 471.3]	0.76
AGPS\|8540	496.45	[505.8, 636.5]	0.97
NLRP2\|55655	854.47	[381.3, 677.5]	0.00

Inter-quartile distance (IQR) is used as a robust measure of dispersion. In the table, IQR(S _n) is the training data’s IQR, followed by the 95-th credible interval, and the posterior predictive P-value. In cases where the P-value is close to 0 or 1, the true test statistic’s distance from the 95-th credible interval can be used to determine the magnitude of the mis-fit.

ISSN: 1471-2105