Phylogenetics: Searching Lab

From EEBedia
Revision as of 19:48, 28 January 2020 by Paul Lewis (Talk | contribs) (Determine NNI rearrangements)

Jump to: navigation, search
Adiantum.png EEB 5349: Phylogenetics
This lab explores different search strategies under the parsimony criterion

Getting started

Log into your account on the Health Center (Xanadu) cluster (ssh Type the following:

srun --qos=general --pty bash

This asks the scheduler to find a node (computer) in the cluster that is currently not busy. It should transfer your session from the head node to a different node. The reason we are using srun today is that some of the analyses we are going to run take more than a few seconds to complete. If all of us ran long jobs on the head node simultaneously, users would notice significant slowdown in response time, which is very annoying to other users. Thus, we will start developing good habits and will use srun to perform our interactive analyses on a node that no one else is currently using.

Once you see the prompt, type

module load paup/4.0a-166

This will make the most recent installed version available to you. Without this line, typing paup may fail to do anything or may start a version of PAUP* that is (perhaps) years old!

Download a copy of the data file

If you do not have the angio35.nex file, you can recreate it as follows:

curl > angio35.nex

The word curl stands for "Copy URL" and the program curl fetches the contents of the specified web site (which in this case is not really a web site, it is just a data file). Without the trailing "> angio35.nex", you would see the contents of the angio35.nex file simply spill across your screen. That ">" symbol followed by a file name causes the output of curl to be captured and redirected to a file named angio35.nex.

Alternatively, you can run the curl command with the -O option as follows:

curl -O

The -O switch tells curl that you wish the file to be saved under its original name (thus making it unnecessary to add the > angio35.nex at the end as we did before).

Create a command file

Now start the nano editor (do not specify a file name because you will be creating a new file) and enter the following text, saving the file as run.nex:


begin paup;
  log file=output.txt start replace;
  execute angio35.nex;

There are at least a couple of advantages to creating little NEXUS files like run.nex. For now, the only advantage is that executing run.nex automatically starts a log file so that you will have a record of what you did. Later, when you get in the habit of putting commands in paup blocks, you will appreciate the fact that your data are separated from the commands that initiate analyses. I have many times opened a data file, forgetting about the embedded paup block that starts a long search and immediately overwrites my previous output files!

Note that because we used the replace keyword in the log command, the file output.txt will be overwritten without warning if it exists. Yes, this is called living dangerously but saves some frustration if a run must be restarted.

Add PAUP commands to this file as you work through the lab today. For each section, add new commands and comment out ones you do not want executed (using [square brackets]). Then, kill PAUP (using the quit command) and restart it using your new run.nex file

Delete all taxa except the first five

Using this command

delete 6-.

will cause PAUP* to ignore all taxa except Ephedrasinica, Gnetum_gnemJS, WelwitschiaJS, Ginkgo_biloba, and Pinus_ellCH.

Exhaustive parsimony search

Use the alltrees command to conduct an exhaustive search under the parsimony criterion (parsimony is the default optimality criterion).

You should add the delete and alltrees file to the paup block in your run.nex file, which should now look like this:


begin paup;
  log file=output.txt start replace;
  execute angio35.nex;
  delete 6-.;
  alltrees fd=barchart;

Start paup, specifying run.nex as the file to execute:

paup run.nex

This analysis should go fast because you now have only 5 taxa. The fd=barchart setting tells PAUP to output a bar chart showing the distribution parsimony scores.

How many separate tree topologies did PAUP* examine? What is the parsimony treelength of the best tree? The worst tree? How many steps separate the best tree from the next best? (consult the bar chart to determine the answer)

Determine NNI rearrangements

Because we performed an exhaustive enumeration, we now know which tree is the globally most parsimonious tree. We are thus guaranteed to never find a better tree were we to start an heuristic search with this tree. Let's do an experiment: perform an NNI heuristic search, starting with the best tree, and have PAUP* save all the trees it encounters in this search. In the end, PAUP* will have 5 trees in memory: the starting tree and the 4 trees corresponding to all possible NNI rearrangements of that starting tree.

Exercise to turn in before the end of lab

Before you start the NNI search, use the showtree command to show you the tree obtained from the exhaustive enumeration.

Draw this tree as an unrooted tree on a piece of paper, abbreviating the taxa as E for Ephedra, P for Pinus, W for Welwitschia, Gg for Gnetum gnemon, and Gb for Ginkgo biloba)

Draw the 4 possible NNI rearrangements (refer to the description of NNI in your lecture notes if you've forgotten) and label them with the tree number from the PAUP* output.

Perform an NNI search

Add an hsearch and describe command to your run.nex file, which should afterwards look like this:


begin paup;
  log file=output.txt start replace;
  execute angio35.nex;
  delete 6-.;
  hsearch start=1 swap=nni nbest=15;
  describe all;

The hsearch command is broken down as follows:

  • start=1 starts the search from the tree currently in memory (i.e., the best tree resulting from your exhaustive search using the parsimony criterion)
  • swap=nni causes the Nearest-Neighbor Interchange (NNI) method to be used for branch swapping
  • nbest=15 saves the 15 best trees found during the search. Thus, were PAUP to examine every possible tree, we would end up saving all of them in memory. The reason this command is needed is that PAUP ordinarily does not save trees that are worse than the best one it has seen thus far. Here, we are interested in seeing the trees that are examined during the course of the search, even if they are not as good as the starting tree.

The describe all command plots the 5 trees currently in memory. The reason we are using the describe command rather than the showtrees command is because we want PAUP to show us the numbers it has assigned to the internal nodes, something that showtrees doesn't do.

Which tree was the original tree? Which trees correspond to NNI rearrangments of which internal edges on the original tree?

Find the most parsimonious tree for all 35 taxa

Modify your run.nex file to conduct a heuristic search on all 35 taxa having the following characteristics:

  • The starting trees are each generated by the stepwise addition method, using random addition of sequences (you will employ the addseq and start keywords for this)
  • Swap using NNI branch swapping (you will employ the swap keyword for this)
  • Reset the nbest option to all because we want to be saving just the best trees, not suboptimal trees (yes, this option is a little confusing).
  • Set the random number seed to 5555 using the rseed option (this determines the sequence of pseudorandom numbers used for the random additions; ordinarily you would not need to set the random number seed, but we will do this here to ensure that we all get the same results)
  • Do 500 replicate searches; each replicate represents an independent search starting from a different random-addition tree (you will use the nreps keyword for this).

Use the following command to get PAUP to list the options for hsearch:

hsearch ?

Remember you can comment out portions of your Nexus file if you don't want to lose them: e.g.,


begin paup;
  log file=output.txt start replace;
  execute angio35.nex;
  [delete 6-.;]
  hsearch [put your new options here];
  [describe all;]
How many tree islands were found? How long did the search take? How many rearrangements were tried?

Conduct a second search with SPR swapping

Construct another heuristic search using SPR branch swapping. Be sure to reset the random number seed to 5555.

How many tree islands were found? What are the scores of the trees in each island? How long did the search take this time? How many rearrangements were tried?

Now conduct a third search with TBR swapping

How many tree islands were found? What are the scores of the trees in each island? How long did the search take this time?How many rearrangements were tried? How many trees are currently in memory (use the treeinfo command)? Has PAUP saved trees from all islands discovered during this search? (Hint: compare "Number of trees retained" to the sum of the "Size" column in the Tree-island profile.) Explain why PAUP saved the number of trees it did.

Wondering about this warning? Multiple hits on islands of unsaved trees may in fact represent different islands When PAUP encounters a new island, it will find all trees composing that particular island in the process of branch swapping. If, in a new search, it encounters any trees already stored in memory, it knows that it has hit an island that it found previously. Note that it would be pointless to continue on this tack, because it will only find all the trees on that island again. For trees retained in memory, PAUP can keep track of which island they belong to (remember that it is possible for trees with the same parsimony score to be in different tree islands!). But for trees that are not retained in memory, PAUP only knows that it has encountered an island of trees having score X; it has no way of finding out how many islands are actually represented amongst the trees having score X.

Of the three types of branch swapping (NNI, SPR, TBR), which is the most thorough? Least thorough? How many TBR rearrangements can PAUP examine on the computer you are using in one second (on average)? Based on this, how long would it take to examine all possible trees if only 20 sequences were included?

Compare NJ with a comparable heuristic search

In this exercise, you will conduct a Neighbor-joining (NJ) analysis using HKY distances and compare this with an heuristic search using the minimum evolution criterion. The goal of this section is to demonstrate that it is possible for heuristic searching to find a better tree than NJ, even using the same optimality criterion.

Please quit PAUP* (if it is still running) and start it again. The purpose of restarting is to return all settings to their default values. You can also simply type the command factory, which resets all of PAUP's settings to their factory defaults.

Create a new file using nano containing the following lines. Note that we are again using the angio35.nex file:


begin paup; 
  execute angio35.nex;
  dset distance=hky objective=me;
  dscores 1; [!*** NJ score above ***]
  hsearch start=1;
  dscores 1; [!*** Heuristic search score above ***]

(The text surrounded by square brackets is a comment, and the initial exclamation point ! tells PAUP that you would like this comment to appear in the output.) Run this file in paup by typing the following at the linux prompt:

paup <filename>

(Of course, replace <filename> with the actual name of the file you just created.)

What is the minimum evolution score for the NJ tree? (scroll down from the beginning of the PAUP* output looking for the phrase "ME-score" right above the point where the comment *** NJ score above *** was printed)
What is the minimum evolution score for the tree found by heuristic search starting with the NJ tree? (Look just above the comment *** Heuristic search score above ***)
What is wrong with this picture? Why is the minimum evolution score of the heuristic search worse than that of the starting tree? (Hint: take a look at the "Heuristic search settings" section of the output.)

Once you have figured out what is going on (ask us for help if you are stumped), fix your paup block and re-execute the file. You may need to get PAUP to help you with the criterion setting; type the following to get PAUP to spit out the current settings, then look for criterion near the top of the list:

set ?

In your reanalysis, you should find that the heuristic search starting with the NJ tree found a better tree using the same optimality criterion (minimum evolution) being used by NJ. Neighbor-joining is very popular, but you should be wary of NJ trees involving large numbers of taxa. This analysis involved 35 taxa; for problem of this size or larger, it is almost certain that NJ will not find the best tree.