Bayesian DNA copy number analysis

BMC Bioinformatics

Table 4 Comparison among the piecewise constant methods on Simulated Chromosomes

method	SSQ	MAD	accuracy	accuracy inside aberrations	accuracy outside aberrations
mBPCR ${\hat{ρ}}^{2}$	1.70	0.00733	0.936	0.992	0.932
mBPCR ${\hat{ρ}}_{1}^{2}$	1.85	0.00781	0.929	0.993	0.920
CBS	1.56	0.00705	0.953	0.985	0.950
CGHseg	5.42	0.00795	0.925	0.885	0.956
HMM	4.47	0.00350	0.993	0.968	0.997
GLAD	4.15	0.00846	0.939	0.930	0.952
BioHMM	5.69	0.003647	0.990	0.949	0.999
Rendersome	19.13	0	0.920	0.289	1

The table shows the comparison of the level estimations obtained using several piecewise constant methods on dataset Simulated Chromosomes. In this comparison, the methods CBS and mBPCR exhibited the lowest SSQ error in the profile estimation and the highest accuracy inside the aberrated regions. On the other hand, HMM, BioHMM and Rendersome had the highest accuracy outside the aberrations, but a high SSQ error. Therefore, the former group of algorithms globally estimated a better profile than the latter. Because of its definition, the MAD error is less informative: it does not take into account if a small number of probes are wrongly estimated, but these probes could correspond to breakpoints or small aberrations.

ISSN: 1471-2105