Table 2.
Parameters in the model are learned using the EM algorithm as discussed below, while hyper-parameters are fixed to the value given
| Parameter | Description | Value | |||
|---|---|---|---|---|---|
| δ | Pseudo counts in Dirichlet prior on π | gN\gT | AA | AB | BB |
| AA | 1e5 | 1e2 | 1e2 | ||
| AB | 1e2 | 1e3 | 1e2 | ||
| BB | 1e1 | 1e1 | 1e3 | ||
| π | Multinomial distribution over joint genotypes | Estimated by EM (M-step) | |||
| Gi | Genotype at position i | Estimated by EM (E-step) | |||
| axi | Number of bases matching the reference genome at position i in genome x∈{N, T} | Observed (JointSNVMix1 only) | |||
| ax:jxi | Indicator that base jx at position i matches reference in genome x∈{N, T} | Latent (JointSNVMix2 only) | |||
| zx:jxi | Indicator that base jx at position i is correctly aligned x∈{N, T} | Latent (JointSNVMix2 only) | |||
| dxi | Depth of coverage at position i in genome x∈{N, T} | Observed | |||
| qx:jxi | Probability that base call is correct in genome x∈{N, T} | Observed (JointSNVMix2 only) | |||
| rx:jxi | Probability that alignment is correct in genome x∈{N, T} | Observed (JointSNVMix2 only) | |||
| μx:gx | Parameter of Binomial distribution for genotype gx in genome x∈{N, T} | Estimated by EM (M-step) | |||
| αx:gx | α parameter in Beta prior distribution on μx:gx | AA | AB | BB | |
| Normal | 1000 | 500 | 2 | ||
| Tumour | 1000 | 500 | 2 | ||
| βx:gx | β parameter in Beta prior distribution on μx:gx | AA | AB | BB | |
| Normal | 2 | 500 | 1000 | ||
| Tumour | 2 | 500 | 1000 | ||