Figure 4

Modelling of the emission probabilities of pHMM U. The model used for the emission probabilities of pHMM U, illustrated by a toy example. Assumed are 3 subtypes (A, B, C), which are assigned to two sources: A to Source 1 (green), B and C to Source 2 (red). Only two nucleotides, G and T, are assumed to exist and is set to (0.1, 0.1). With probability η the right-hand part of the model is chosen, and with probability 1-η the left-hand one. If the left-hand part is chosen, then with probability Source 1 is then chosen, and with probability Source 2. At the bottom, the generated emission probabilities are given for the different paths the model can take. In case the right-hand part is chosen, a Dirichlet distribution with parameter is taken for the generation of the emission probabilities. The emission probabilities of pHMM U are estimated by averaging over all possible emission probabilities, weighting them with their respective probabilities. That is, assuming η = 0.05, we obtain as estimate for the emission probabilities.