celinescornavacca
commented
3 years ago In their paper, they use this parameters in the simulations
\lambda - \nu ~ exp(0.1) with mean 10 \tau_0 ~ exp(10) with mean 0.1 \nu / \lambda ~ Beta distribution with alpha = 1 and beta = 1 (same as uniform distribution between 0 and 1) \gamma ~ Uniform distribution (between 0 and 1)
As I said in a slack message, this process gives Yule-type networks, that act as the Yule-type trees, see example below for trees:
Once they start having a certain amount of nodes, things get crazy and they start speciating and hybridising a lot. It is not the focus of our paper to correctly simulate networks (we are not in a Bayesian framework, we do not need priors) and this is the best we can for now (for the next paper, we will do better, working on it). So we can simply throw the networks we do not like as done in line 143-147 of this simulator, which only simulate networks and not trees and sequences.
RandomNetwork.txt (in txt because py is not supported)
Sarah, you will find the parameters they use in the paper on lines 8-17, you can copy and paste this in your simulator and add a line similar to line 143, and you should be good.
lutteropp
stamatak
We are going to simulate our networks under the birth-hybridization process of Zhang et al 2018. In this process, we have a speciation rate l (lambda), a hybridisation rate v (nu) and the time we run the process t_0.
From their paper: "The simulator starts from the time of origin (t_0) with one species. A species splits into two (speciation) with rate l, and two species merge into one (hybridization) with rate v. At the moment of k branches, the total rate of change is r_tot = k l + choose (k 2) v. We generate a waiting time ~exp(r_tot) and a random variable u ~U(0,1), If u < k*l=r_tot, we randomly select a branch to split; otherwise, we randomly select two branches to join, and generate an inheritance probability c ~U(0,1). The simulator stops at time t_0."
Zhang et al. - 2018 - Bayesian Inference of Species Networks from Multil.pdf
No transfer from now, because branches with no length cannot be recovered.
Given such generated network, we extract a random tree, using the inheritance probabilities at the reticulated nodes: for each reticulated node x, we draw a number y ~ U(0,1) and we choose the first parent p_x if y < inheritanceProb of the edge p_x,x. If we do this,we have atree but not a binary one. Then we clean the tree from the dead nodes and 1-indegree 1-outdegree nodes, as Sarah explained in her document, and we obtain a binary tree. On this binary tree, we simulate N sequences with Seq-Gen-1.3.4 (currently under this model -mHKY -t3.0 -f0.3,0.2,0.2,0.3 this can be changed). We do this K times, obtaining an alignment of length K*N in which each group of N sites has its tree.