I am doing some tests with covid data using genomes from the gisaid database, but frequently, after the analysis ends I get a tree with a lot of 0 values in the branches corresponding to the posterior probability. How can I improve them? There is a practical method to check if the dataset is ok before running BEAST?
Hi Pedro Jeronimo, I'm afraid that there are no "silver bullets" in here... Beast trees are very complex constructs, and the lack of posterior support may come from several parts of the process. It would be helpful if you share the main parameters of your setup as well as some basic info about your dataset (sizes, S, singletons, ...).
Usually, low supports mean that you have several equally likely solutions, which might be resulted from a highly homogeneous dataset or MCMC too short to explore the complexity of your sampling space.
Make sure your basic stuff is ok. I recommend you try your setup in some example data and see if it is working as expected. Example data are a good start because you do have a clear output expectation. Then you run it on a sample of your data (only a few samples or representative parts of the genomes you have).
After you make sure your setup is ok, and your sequences are not too similar, or MCMC is not too short, I would think about the tree prior (model, parameters, and hyperparameters); and then the nucleotide substitution model (more parameters they have more suitable for tiny differences - theoretically). Most of the time, the problems we face are really simple. Make sure the basic is fine before looking out for more complex stuff.
Hope it helps
Good luck
Wilson
You can improve it by improving the alignments used in creating the tree.
- Badges
- Science topic
- Similar topics
- Analytical Chemistry
- Extraction