hsegHMM: hidden Markov model-based allele-specific copy number alteration analysis accounting for hypersegmentation

Table 2 Summary of simulation studies with hsegHMM-N and hsegHMM-T models based on 500 simulated datasets

Simulation 1 and Simulation 2 are the t-distribution-based and the normal-mixture-based studies. Each dataset consists of 10,000 observations of logR and logOR. Est is average estimates from 500 datasets ;ψ is the ploidy, α is the tumor purity; κ² is the variance component of logR in hsegHMM-T; V(W) and τ² are the variance of logR and logOR in both models, respectively; SEs indicates the Monte-Carlo standard errors calculated from 500 datasets; SE_H indicates the average asymptotic standard errors of estimates based on the Hessian matrices
∗ the average asymptotic standard errors based on the hsegHMM-N model are reported based on 486 datasets where 2.8% of 500 datasets cannot produce invertable Hessian matrices due to numerical problems
^aV(W)=E(V(W|u))+V(E(W|u))\(=\kappa ^{2}\times \frac {v}{v-2}\)
^bthe asymptotic standard error of V(W) with the hsegHMM-T is calculated by using the Delta method
△ The distribution of the ploidy estimates is skewed so the SE _s of the ploidy appears to be larger than SE _H. Using the scaled MAD (median absolute deviation) gives a closer value (0.008) to SE _H; \(MAD=1.4826\times med(|\widehat {\theta }_{m}-\widehat {\theta }_{med}|)\), where \(\widehat {\theta }_{m}\) is the estimate for the m^th dataset and \(\widehat {\theta }_{med}\) is the median calculated from 500 simulated datasets

ISSN: 1471-2105