 Hello and welcome to today's presentation on an improved view of influenza evolution with Colry. I am Ugnar Stoltz, a PhD student at Tanya Stadler's computational evolution group, and will be introducing you to a new framework for phylogenetic inference on viral reassortment networks. It is based on a recent publication by Nicola Mueller, Tim Vaughan, Tanya Stadler, Gita's Dudas and me, which focused on reassortment patterns of human influenza viruses. First, I would like to introduce you to the concept of viral phylogenetic networks in the context of reassortment. Here we see a blue infected host. Going forward in time, they come into contact and infect the green person, who subsequently infects the red person. Then we can obtain viral sequences from all these individuals. In practice, this is usually where we start. That is, we observe infected individuals and collect viral sequencing data from them. Since viruses typically accumulate mutations rapidly, we can construct their phylogenetic tree using a coalescent model, starting at the samples and going backwards in time. Then, coalescence of two viral lineages can be viewed as approximation of the transmission event, shown here. However, these phylogenetic trees cannot fully capture the evolution of segmented RNA viruses, for example, influenza or LASA virus. Their genome is comprised of several distinct RNA molecules called segments, and during co-infection of the host cell, they are able to exchange these genetic segments during a process called reassortment. Here you see two parent viruses with three segments each, co-infecting a single host cell. During the viral replication cycle, each of these three shown segments is replicated separately before being repackaged into a new viral particle. Since segments of both parent viruses are being replicated, they might mix during the full genome assembly and thus create a new hybrid progeny. In our toy example, imagine that the red person came into contact with both green and blue hosts. Then, its cells carry two distinct viral variants and through reassortment produce a novel viral strain. As you can see, in order to depict the situation, we had to modify the tree by connecting blue and red lineages by a vertical line shown here, and thus having a network. Therefore, having our toy example in mind, we can say that evolutionary analysis of reassorting viruses requires networks, not trees. In order to infer such networks, we need an evolutionary model. We have extended the Kingman's coalescent process for trees to include reassortment of viral lineages. In this phylogenetic network model, ancestral lineages carry genomic segments, here shown in three colors, of which only a subset may be ancestral to sampled viral genomes. Here, they are depicted in solid lines, which show the embedded trees of the segments. Network lineages coalesce with each other backwards in time, as shown here by the coalescent events. At reassortment events, a random subset of segments follow each new lineage. For example, here, the green segment follows left parent, while the other two segments follow the right parent. This allows us to infer reassortment networks and the embedded segment trees for sampled sequencing data. This model is implemented in a coalescent package, which is available as an add-on for phylogenetic software BS2. Taking viral sequence alignments as input, it provides the posterior distribution of coalescent, reassortment, and evolutionary rates, as well as network topologies. Next, I will focus on an example when Coal-Ree was applied to several strains of human influenza virus. As mentioned, influenza is a segmented virus and is comprised of eight genomic segments. Influenza viruses have four types, out of which type A and B are known to cause seasonal epidemics. It can be further differentiated by its surface proteins, coded by hemagglutinin and neuraminidase segments, shown here in blue and yellow. We have run Coal-Ree analysis on genetic sequences from four influenza A strains and one influenza B strain. On the left, you can see an example, maxim click credibility network that summarizes the posterior distribution of networks for pandemic 2009-like influenza AH1 and 1 strain. The vertical color lines show the reassortment events between the network lineages. On the right, the posterior distribution of reassortment rates for 10 sub-samples of each dataset are shown. As you can see, significant differences in reassortment rate can be observed between influenza viruses. Next, we compare the uncertainty in node-hide estimates obtained by Coal-Ree or the model which assumes that each segment calls for an independent realization of the same coalescent process and thus does not account for the assortment. We have quantified the uncertainty as a relative width of 95% highest posterior density intervals and found that Coal-Ree reduces this uncertainty by at least a half. Posterior click credibility support was also higher for Coal-Ree. Therefore, accounting for reassortment events reduces bias in influenza inference. Lastly, the posterior distribution of rates obtained by Coal-Ree or assuming independent segments were significantly different. Assuming independent segments and therefore not accounting for reassortment leads to lower effective population size seen here and does increased coalescent rates which are inversely proportional. In addition, not accounting for reassortment also leads to higher evolutionary rates. Next, I will show you a quick introduction on how to set up your own inference with Coal-Ree. I will assume that you have the required resources installed. You may find their location in the description of this video. In order to run Coal-Ree analysis in v2, you'll have to create an XML file which includes the information about the data you're going to use, your model of choice and parameter prior distributions. This XML can be easily created in beauty a user interface for v2. There are several tabs. In the first one, you have to import sequence alignments for all segments that you are going to analyze. Here you can see HA and NA. In the next tab, you can automatically generate the dating of your samples. Next, choose the substitution model and its parameters that you would like to use for this analysis. Specify the clock rate and choose your prior distributions. Lastly, select how long you would like your MCMC chain to run. When you are done, save the XML file. Use this XML file as input to beast and run the analysis. Once it is done, in your output directory, you can find several log files, two segmentary log files, one log file for the posterior distribution of networks, which invents the segmentaries, and finally a log file for the summary statistics of the network as well as samples on the posterior distribution of its parameters. This log file can be viewed with Tracer application that is provided together with beast and allows to inspect the locked summary statistics and posterior distributions. Finally, you can summarize the posterior distribution of networks into one maximum-quade credibility network. This can again be done from beauty application. Select File, then Launch Apps. Then in the App Launcher, choose Colory and Reassortment Network Summarizer. Once you launch this app, provide your log files and analyze them. After it is completed, you will have one more log file which contains the maximum-quade credibility network. It can be viewed on the online application ic3.org. Simply drag and drop the network file in here and it is displayed. The reassortment events are shown with one parent image in dashed line. In the Style section, you can choose to see the 95% highest posterior density intervals for the node heights of the network. If you would like to know more about setting up the Colory analysis in Beast2, visit tamingthebeast.org website where you can go through the detailed tutorial with the same data I have used today. It walks you through the necessary programs that you have to install, setting up the XML and analyzing the outputted log files as well as how to summarize the networks and obtain the maximum-quade credibility network. The tutorial concludes my talk today. Thank you all for joining.