Skip to content
Permalink
Branch: master
Find file Copy path
Fetching contributors…
Cannot retrieve contributors at this time
68 lines (33 sloc) 2.89 KB

Unit 14 - Substitution Models

First lets get everyone on the same page with their .bashrc. Log onto Hrothgar.

IMPORTANT: IF YOU HAVE A CUSTOM .bashrc DO NOT DO THIS. IT MAY WIPE OUT IMPORTANT FEATURES SPECIFIC TO YOUR ANALYSES

cp /home/roplatt/.bashrc ~/.bashrc

Now make sure you have the following programs installed.

And here is today's data (link). Save it as bar1ML_aln.fas

Everything we do in this unit will be on your local computer.


Models

All of the following analyses will be done in MEGA6 untill notified.

  1. Calculate the number of substitutions (No. of differnces) in a pairwise alignment of the bar1ML_aln.fas data

  2. What is the difference between pairwise, partial, and complete deletion?

  3. Calculate the precent of substitutions (p-distance) in a pairwise alignment of the bar1ML_aln.fas data.

  4. Calculate the substitution matrix for the data using the Jukes-Cantor model. This is the most simplistic substitution model. How does this compare with your data. Look at it. Also, take note of the likelihood score.

  5. Recalculate the substitution matrix for the data using the most parameter rich model available (GTR+I+G). Note the likelihood score.

  6. Recalculate the substitution matrix for the data using the remaining models (JC, K2P, T3P, HKY, TN). Note the likelihood scores of each.

  7. Using something like Excel, make a graph showing the change in likelihood score for each model (put the models in order of parameter richness). Notice a trend?

  8. Now go back and compute the overall mean distance using JC69, K80, & T3P.

You should notice that the model really does affect the distance. At a certain point adding parameters does appreciably alter your distance values

How do we choose which model is best in a statistically justifiable way?


Model Choice

  1. Use MEGA6 to test for the best fit model. Which is it? How does this compare with your graph of likelihood scores? How many models were tested.

  2. Use jModelTest to choose the best-fit model. Run this with minimal substitution schemes. How many models were tested? How many models are available to be tested?

jModelTest can test >1,600 models. Most downstream phylogenetic programs cannot use most of those. Other programs like partitionFinder can limit the models tested to thos capable of being implemented in major phylogenetic programs.


Updated: 4:00PM - 28 Oct 2015

Molecular Systematics Zool6505

You can’t perform that action at this time.