Chunk #15 — MATERIALS AND METHODS — Statistical model — QuantiSNP: an Objective Bayes Hidden-Markov Model — Objective learning, expectation maximization(EM) and the Viterbi algorithm
Our model will be calibrated to a user-defined specificity (false positive) rate of excursions out of the normal (copy number = 2) state, however, we wish to restrict the number of prior parameters which need to be tuned in this manner. Hence, we choose to estimate most of the hyperparameters, λ = {τ, α, β, m}, via maximum marginal likelihood techniques to a reference dataset obtained from chromosome X multiple copy cell lines (20), 9 with the remaining (user specified) free parameter L in Equation (1) to be calibrated against Type I error, as described below.
