Fig. 6

Overview of the CNRein algorithm. a The probability of read depths R and BAF values B is determined by the variance levels \(\varvec{\Sigma }^{\textbf{R}}\) and \(\varvec{\Sigma }^{\textbf{R}}\) as well as the copy number profile P. The probability of copy number profiles P then comes from the model parameters \(\theta\). b A potential copy number profile \(g([ c^1, c^2, c^3 ])\) generated by applying mutations \(c^1\), \(c^2\), and \(c^3\) to a normal cell is shown. Specifically, the copy numbers for haplotype 1 and haplotype 2 are plotted and for each CNA tuple, the haplotype number, the start position, the end position, and the value are shown in the copy number profile. c In CNRein’s evolutionary model, a copy number profile is generated by the application of a series of CNAs such that the probability of each new CNA depends on the previous copy number profile. In this example, CNRein generates the copy number profile in panel b by sequentially applying the mutations \(c^1\), \(c^2\), and \(c^3\) to a normal cell. d The probability of CNA events are determined by a neural network. \(g([c^1, c^2])\) is the copy number profile generated by applying the two CNA tuples \(c^1\) and \(c^2\) to the normal cell. Then, the probability of the components of CNA tuple \(c^3\) is generated. Ultimately, \(\Pr (c^3 \mid g([c^1, c^2]), \theta )\) is the probability of the next CNA \(c^3\) given that CNAs \(c^1\) and \(c^2\) have already been applied