Recently in Problems and projects Category

I've had several questions about question #3 on project #3. All of them are something like "How do I answer that?" In the interest of sharing the same advice with everyone. Here's roughly how I've responded each time I was asked:

Read the papers. Think about the results of your analyses and what they tell you about the underlying evolutionary processes that produced the patterns you see. Ask yourself "Are the results of my analyses consistent with what the authors claim?" Once you can answer that question and can explain your answer, you've got the answer to question 3.

Hi all,

I will be in my office on Wednesday from 1-6 and Thursday morning 8:30-12.  If you need help and these times do not work for you, please email me.

Cheers, Kathryn

As you may recall, I'll be grading Project #3 -- not Kathryn. As you'll also recall, I'm leaving town on Saturday, so if you don't get Project #3 handed in on time, I'll have no choice but to give you an incomplete. I also plan to leave campus almost immediately after lecture on Friday, and I may or may not be able to read projects that are e-mailed to me.

Bottom line? If you want to receive a letter grade rather than an incomplete:

1. Hand in a printed copy of Project #3 on Friday.
2. Hand it in during lecture on Friday.

If you send me an e-mail copy or leave a hard copy after lecture, I will grade it. But the chance that I'll get it graded in time to assign you a letter grade for the course is not distinguishable from zero.

If you tried to download the second paper for Project #3 (the one from Molecular Ecology), you will have found that the link was broken. I've fixed it now, so you should be able to get it without any trouble.

Happy Thanksgiving!

I've just posted Project #3 to the course web site. You'll find a link to the assignment and the associated data set on the detail page for tomorrow's lecture. I'll need to introduce you to one final software package, Arlequin. The current version is only available on Windows, which is the version I'll show. But I think that the earlier version that runs on Macs will work for this project too. The interface is very similar. I may not have a chance to try the Mac version before lecture tomorrow, but I'll do my best to try it with these data by late on Monday, and I'll let you know whether you'll be able to run it on a Mac.

I've heard from a few more people about problems. If you get this message and have time before lecture this morning, please try this and let me know what happens.

I've had reports from at least two people that WinQTLCart freezes on them when they try to do an analysis, either IM or CIM. I am investigating and trying to diagnose the problem. In the meantime, if you haven't downloaded WinQTLCart and the data yet, please give it a try and let me know whether or not it works for you and what operating system you're running.

I think I know a workaround, but it's one I'd prefer to avoid using unless we have to. I'd rather fix the problem.

As promised (threatened?), I've posted the QTL problem. I'll need to show you two or three more things about  QTL Cartographer before you can do the whole analysis, but if you'd like to download the data and make sure you can read it into QTL Cartographer, everything is available from the detail page for tomorrow's lecture.

Oh, one word of warning. Each analysis for composite interval mapping is going to take a long time. I started an analysis of the %N at 25°C data this morning around 10:00am, and it's still running. I mention this as a warning. You're going to want to start your analyses fairly soon, and you'll probably want to set them to run overnight. If CIL takes inordinately long, which I'll know by lecture tomorrow, we'll settle for an interval mapping analysis instead.

For project 2 I'd like you to write out a couple of sentences/paragraphs answering each question.  After that, I'd like you to attach the output from WinBUGS and then your WinBUGS code WITHOUT the data.  This can be emailed to me to save paper.

This is the suggestion I've been giving out, which seems to work well.
1)  Start with the basic code Kent gave and modify it to fit the Mimulus dataset.
2)  Starting with question 1, figure out which variables you need to solve for and then find the equations using terms you (well, technically Kent in this case) have already solved for, then write out the equations in WinBUGS format.
3)  Repeat Step 2 for each problem SEQUENTIALLY.
4)  Send me your code before you run it.  Depending on how you wrote it, there are a couple of different methods to loading the data.

I will be available on Wednesday between 8:30 and 5 (with the exception of 12-12:30 for lunch).  Please stop by my office in Pharm/Bio if you need help.

I've posted a version of the Mimulus data without the last column, so you can try your hand with the code that's already available (before you start worrying about the regression on fitness).

Hello all,

My available schedule for this week is as follows:
Monday 4-6pm
Tuesday 8am-2pm
Thursday 8-10am
If people can not make ANY of those times and still want help, please email me.

I've just posted the data and problem description for Project #2. You'll find the link on the lecture detail page for tomorrow.

As promised (threatened?), you're going to have to fire up WinBUGS again for the solution. That's the bad news. The good news is that you can take the full-sib, half-sib code that I posted when we were talking about variance partitioning on Monday, change the numbers for the number of individual, the number of sires, and the number of dams, and find yourself most of the way to the solution.

Strictly speaking, you wouldn't have to fire up WinBUGS. You could fire up R, SAS, SPPS, or another statistical package and estimate variance components there. But if you do, you'll also need to figure out how to bootstrap over full-sib families to get confidence intervals for your results. While that's possible, it may not be very easy, and I'm afraid you'll be on your own if you try it that way. We won't be able to offer any advice. (Now if you wanted to combine mcmcsamp() with lmer() results in R, I could help you, but (a) you'd need to understand what mcmcsamp() does and you'd need to be familiar enough with R to write your own code to analyze the results it gives you.)

It's taking me longer to put Project #2 together than I had planned. (I'm trying to make sure that the data make sense and aren't too difficult to make sense of.) As a result, instead of assigning it tomorrow, I won't assign it until Friday, and it won't be due until a week from Monday (the 3rd of November). That gives you two weekends to work on it, including Halloween. It also means that we will have covered all of the evolutionary quantitative genetics stuff that's going to be part of the project before it's assigned.

Hello everyone!

If you are planning to come meet with me on Wednesday, please be advised that I will be in the fishbowl (across from my office) for help from 9 to 11 and 3 to 5 (ish).  If you know that you are coming to meet with me at a specific time, please email me to let me know so that I can try to avoid backups!

Cheers
Kathryn

I'll give Kathryn a copy of my (WinBUGS) solution to the bonus question later today or tomorrow. So she should be able to give you some more hints after that. In the meantime, I've posted a WinBUGS-friendly version of the data to the course web site (data.txt). Here are the variables you'll find in the data set:

• n.one -- The number of replicate populations produced in the one-generation experiment (100).
  
• n.one.min -- The mininum possible number of alleles in the population (because the sample contains at least that many copies, 34).
• n.time -- The number of generations for the time series (50).
• n.time.min -- The minimum possible number of alleles in any of the populations (because the sample contains at least that many copies, 52).
• one -- The replicate population data for the one-generation experiment.
• time -- The data for the time-series experiment.

Notice that the data has been converted to integers for you. I'll post another hint tomorrow or Wednesday (unless I get questions sooner).

I've posted the description of Problem #3 and two (simulated) data files associated with it. Unlike the problems/projects so far the bonus question associated with this problem is very difficult. I'll be astonished (and very impressed) if anyone is able to solve it without help, but I'll be providing hints over the next week or so for anyone who's interested.

Kathryn is leaving for the airport in five minutes. If you haven't sent her your problem #2 answers yet, she asks that you cut and paste them into the body of your e-mail. She'll be able to check e-mail when she's in Antanarivo Sunday and Monday, but Internet connections are very slow, and an attachment may not be downloadable. It might be a good idea to send me a copy of your answers too, just to make sure that one of us has a copy.

I've heard from a couple of people who are having difficulty getting started on Problem #2. Here's the hint I gave them, and I thought I'd pass it along to everyone.

1. Remember the kind of selection we've been focusing on, namely viability selection.
2. Remember the definition of viability selection, namely the probability of zygote to adult.
3. Remember the assumptions in the problem statement, namely that 90% of those diagnosed with severe malaria dia and that anyone who is going to get severe malaria gets it before they have a chance to reproduce.
   

If you work through the implication of those three observations, you should be able to estimate fitnesses for each of the three genotypes.

To answer the second question you're going to need to think of a way to compare fitnesses. You don't have to use WinBUGS to answer that question. You could do a chi-squared contingency test, rather than using WinBUGS. But to get full credit for your answer on this question, you'll need to do one or the other.

If you've already finished Project #1 and are looking for something to do, you'll find a link to Problem #2 on the Detecting viability selection lecture page. There isn't a link to a dataset, because the data you'll need are included in the problem statement.

I've just posted the spreadsheet with consolidated results from the structure analysis. You can follow the highlighted link in this post, or you can find the link along with all of the other project materials on the lecture detail page for 10 September, when Project #1 was assigned.

Hi All,

I am still waiting on the last few people to send me their Structure data.  I teach from 2:30-6:30 this afternoon, so if everyone gets me data, the spreadsheet will go out this afternoon.  Otherwise I will check my email at 6:30 and add whatever data is left.  I will be gone the rest of the evening, so if you don't have your data to me by 6:30, it will be considered late.

Cheers
Kathryn

I've heard from a couple of people that they're getting an error message when they try to run Structure: "Could not create Java virtual machine." I'm not sure what's causing the problem, but here's one thing to try if you're running into that problem.

• Reinstall Structure, and when you get to the "Install Type" screen, check the radio button for "Custom".
• Click "Next", and make sure all of the check boxes are checked, especially the one labeled "Jre".

Let me know if Structure still doesn't work after that. We'll figure out another way to handle the problem.

There appears to be a bug in the theta=0 model in Hickory. I'll see if I can find a workaround, but unless you hear otherwise, plan to do only the full model and f=0 model in Hickory.

We went over how to run Structure pretty quickly on Friday, and I've heard from a couple of people who are having some problems. Here's a list of instructions so it's all in the same place.

• Double click on the Structure icon and start the program
• From the  "File" menu, select "New project".You'll get a "Project Wizard" box labeled "Step 1 of 4".
• Give your project a name, e.g., "humans" (without the double quotes)
• Select the directory where you want the results to be stored, probably the directory where you downloaded the file humans.stru.
• Select the humans.stru file and click "Next >>". You'll get a "Project Wizard" box labeled "Step 2 of 4"
• There are 263 individuals in the data set, the ploidy is 2 (diploid), the number of loci is 100, and the missing data value is -9. Enter those values in the respective boxes and click "Next >>". You'll get a "Project Wizard" box labeled "Step 3 of 4".
• Check the box labeled "Row of marker names", and click "Next >>". You'll get a "Project Wizard" box labeled "Step 4 of 4".
• Check the box labeled "Putative population origin for each individual", check the box "Other extra columns", enter 2 in the box labeled, "Number of extra columns",and click "Next >>". You'll get a "Confirmation" box. Click "Proceed" and you sho uld have the data appearing as a spreadsheet in Structure.

At this point you have the data loaded into Structure and you're (almost) ready to begin the analysis. First, you have to set up a "Parameter set".

• To do that select "Parameter Set -> New". You'll get a "New Parameter Set" box popping up.
  
• We want a burnin of 10,000 iterations and a sample of 50,000 iterations. Enter those numbers in the respective boxes and click "Ok".
• You'll get an "Input" box popping up asking you for the name of your parameter set. You can pick anything you want. I usually pick "default". Enter the name you want in the box and hit "Ok". The box will disappear, and you'll see in the list of things on the left side of the screen a new Parameter set with the name that you just provided.

Now you're ready to run the analysis.

• Select "Project -> Start a job", and a "Structure Scheduler" box will pop up.
• Highlight the parameter set you created, set K from 2 to 10 by entering the numbers in the appropriate boxes. Leave the number of iterations set at 1.
  
• Click "Start". The "Strucutre Scheduler" box should disappear, a "Structure Job Log" box should appear, and you'll start seeing numbers appearing in the black box at the bottom of the screen.

It will probably take 10-12 hours for the analysis to finish. If you need to stop it along the way, select "Project -> Kill Running Job". You can pick up where you left off by selection "File -> Open Project", setting the "Structure Scheduler" to start with the K you had when you started the analysis and clicking "Start".

Since we need results from this analysis by Tuesday morning so that we can post the results, it would be a very good idea to see if you can get the analysis running today so that we can go over any problems tomorrow, even if you don't finish the analysis tonight.

I've just uploaded the spreadsheet for the Structure part of the assignment to the website. Here's the link. There will also be a link to it from the assignment page shortly. 

I just uploaded a new version of the NEXUS file for GDA. None of the data have changed. So if you've already run analyses with GDA, you don't need to run them again. But the new file has an addition that you'll find helpful in answering the second question. I'll describe the addition in lecture today.

Hi all,

For any of you who are thinking to track me down on Monday morning, I've got some bad news.  I'm have a meeting and then a doctor's appointment so probably won't be around until about 10:30. 

Kathryn

Several people have asked for clarification on the bonus question. Let me see if I can state the question in a way that makes my intent clearer.

What frequency of homozygotes would you expect in the population if the allele frequencies you just estimated are characteristic of the population and genotypres are in Hardy-Weinberg proportions? Suppose that you didn't know there was a null allele, i.e., you assumed that those individuals who are homozygous null were bad samples that could be ignored. What frequency of "homozygotes", i.e., phenotypes you'd score as homozygous even if they really aren't, would you expect to find in your sample?

Let me know if that doesn't help, and I'll try again.

If you're having trouble with Problem #1, you're not alone. I've spoken with many of you and shared some hints, so I thought I'd write them down and pass them along.

First, if I told you there were five alleles segregating in this population (A₀, A₁₁₃, A₁₁₉, A₁₂₃, and A_u) with frequencies p₀, p₁₁₃, p₁₁₉, p₁₂₃, and p_u I'll bet you could figure out how to calculate the frequency of each offspring category in this table, right?

So do that and figure out how the phenotype categories in the problem correspond to the genotype frequencies in this table. (You'll have to do this for the other maternal genotype too, obviously.)

That's the hard part, and that's the part I'm really interested in. That's biology (or at least genetics). The rest is translating the formulas you have in that table into WinBUGS.

For that, stare at the code we used for estimating allele frequencies in the ABO blood group system a bit, try your hand at a solution, and if you get stuck, stop by or e-mail Kathryn or me for some additional hints.

Just so we're all on the same page, I thought it would be a good idea to let you know two things about what and when we're expecting to receive your answers to Problem #1.

• We're expecting you to turn in printed copies of your answers, including the WinBUGS code you used to solve the second problem.
• We'll ask that you turn in your answers at the beginning of class on Monday. Talk with Kathryn if you think you're going to have trouble making that deadline.

With respect to late assignments, I should remind you that once you start to slip behind, you're liable to get farther and farther behind. The final deadline, for Project #3 is an absolutely firm deadline. I leave for New Zealand the next day, and to be sure that I'm able to get grades posted in PeopleSoft, we'll have to have everything graded by the 10th of December.¹ If Project #3 comes in late, you will receive an incomplete for the course.

If you haven't tried using WinBUGS yet, I encourage you to do so soon. It's not the easiest program to learn, and in addition to learning how to use it, you have a problem that you'll need to solve with it. That means you'll need to write your own WinBUGS code. I'll be amazed if you manage to solve the problem without asking for help, and I can't promise that Kathryn and I will be able to help much if you wait until next weekend to give it a shot.

Oh, you'll find some hints in the comments to the Problem #1 update that you may find helpful.

Kathryn pointed out that I should make it explicit that there are no A₀A₁₂₃ offspring from the first mother. I've updated the assignment accordingly.

You'll find Problem #1 posted in both HTML and PDF formats on the page associated with today's lecture: http://darwin.eeb.uconn.edu/eeb348/lecture.php?rl_id=396. If you have questions about the problem, you may post them here if you'd like. Keep in mind that you won't get a notice when your question is answered, so you'll need to check back occasionally to see if we've answered.