Genome 560 Spring 2011

Genome 560

Statistical Genomics (actually it should be called Statistics for Genomicists)

Spring, 2011

Table of contents:

What this course is and isn't
News about the course
A rough syllabus (to be improved)
Lecture PDFs
Audio recordings of lectures
Books
The R language
R in this course (and R exercises to be done in class)
Homework assignments
The lecturer

What this course is and isn't

I inherited this course, and I'm not entirely sure what it is supposed to be. Its title implies that it will tell you about advanced statistical methods in genomics. It won't. It will not tell you about:

probabilities of sequence motifs
distributions of restriction sites
ways to test differences in base composition
Markov chain Monte Carlo methods
Hidden Markov Models
EM algorithms
Phylogeny inference methods
coalescents
statistical aspects of population genetics
pedigree analysis
tracts of identity by descent
junctions approaches to inbreeding
genome rearrangement statistics

and all sorts of other advanced genomic topics.

What is does seem to be is a statistics refresher, for genomics students who have had a “cookbook” statistics course. For those who have not had any statistics, it is a minimal introduction. But I do not think that it can be considered an adequate substitute for a real statistics course. Genome Sciences graduate students ought to have a full statistics course.

There are three levels of statistics course at most universities. Here is a slightly tongue-in-cheek summary of each of them:

“Cookbook” courses where large numbers of unwilling students are pressured into learning statistics by rote, usually taught by a teacher who is not there voluntarily either. A recipe for disaster. The convoy moves at the speed of its slowest ship, so days are spent on what a histogram is. You can detect whether a course is a cookbook course by looking in the Index of its textbook. If neither “likelihood” or “moment-generating function” are mentioned, that is a cookbook course.
An upper-undergraduate statistics course that tries to explain some of the logic. These are probably the level you would want. Their textbooks do have index entries for “likelihood” but usually not for “moment-generating function”. One difficulty is that the examples used in the course are probably not similar to the problems we encounter in genomics, but may instead be motivated by other fields such as engineering or medical clinical trials. Here at UW some this level of course is represented by Statistics / Math 390. The Statistics 340-341-342 sequence is at that level, but is a more complete mathematical statistics sequence (see below).
A full graduate-level mathematical statistics course. The textbook index mentions both “likelihood” and “moment-generating function”. These sequences are generally full-year courses intended to get a budding mathematical statistician up to speed. Some of the methods taught are not methods of data analysis but methods of proving theorems (such as how to prove convergence of a quantity to a particular distribution). This may even leave you feeling wise and powerful but unable to analyze any actual data.

Here is what you need to do in addition to this course: take a real statistics course. Here is what you should not do after taking this course: tell your thesis committee “oh yes, I've had statistics!” If you do I will come after you and personally lash you with a wet noodle.

News about the course

These news items have the newest ones last.

The course meets in S110 Foege Building, the seminar room on the 1st floor of Foege (its door is across from the white phone in the hall). It meets at 9:00-10:20 Tuesdays and Thursdays, May 2 onwards.

There is a course mailing list. Registered students are on it automatically. For others, to join it or read past postings, go to this link. It requires a UW login, and past postings can be read only once you are a list member. Members can post to the mailing list. I will try to keep anyone from abusing the list and sending spam or really off-topic stuff.
The course will be graded on homework assignments assigned here each Thursday, to be turned in the following Tuesday. They will make use of the R package.
The R statistical package will be used in homeworks. Students should load it onto their own laptop and bring the laptop to class, as we will do exercises, group learning, and group confusion the last 20-30 minutes of each lecture. R is free and can be downloaded from the Cran-R web page here. A brief introduction document to R is available there as a PDF. Here is another good set of pages that introduces R. Two good quick example sheets were made up by Josh Akey two years ago. They are
An R tutorial
and An R descriptive statistics tutorial.
They show you the commands to do some relevant things, but do not show you what appears after you type in the commands. I recommend that you try typing in these commands yourself -- it will be quite helpful.
The grades were submitted (on time). I have posted a histogram of the grade point totals (out of 100) with some indication of which ones got which grades. The histogram is a PDF that can be found here.
Later today I hope to have the homework papers returned to the students' mailboxes in Foege. If you do not have a mailbox there, please email me as to where to send the homework papers.

A rough syllabus (to be improved)

(May 3) Probability. Stochastic processes (coins, phone calls, normals) Distributions (uniform, binomial, hypergeometric, geometric, exponential, Poisson, normal, lognormal)
(May 5) Distributions, cont'd. Histograms, etc. Quantiles, distributions of functions of (multiples, averages, sums, sums of squares, differences) Practice R
(May 10) Confidence intervals, t-test, experimental design, tests
(May 12) Chi-squares, contingency tables
(May 17) Regression, curve fitting
(May 19) ANOVA, F-tests
(May 24) Bayesian inference, likelihood
(May 26) Jackknife, bootstrap, permutation tests, cross-validation
(May 31) Multiple testing: Bonferroni, modifications of, FDR
(June 2) Principal (no, not “principle”) components and SVD

Lecture PDFs

The lecture PDFs will be posted here. One from this year and others from last year are linked here now (the latter are marked here as "old").

Audio recordings of lectures

The lectures will be recorded and made available as WMA and as MP3 files here. They will be recorded at medium quality, the files about 10 Mb in size. Each file has a name which is the date on which it was recorded, such as 20110503.WMA or 20110503.mp3.

May 3	May 5	May 10	May 12	May 17	May 19	May 24	May 26	May 31	June 2
20110503.WMA 20110503.mp3	20110505.WMA 20110505.mp3	20110510.WMA 20110510.mp3	20110512.WMA 20110512.mp3	20110517.WMA 20110517.mp3	20110519.WMA 20110519.mp3	20110524.WMA 20110524.mp3	20110526.WMA 20110526.mp3	20110531.WMA 20110531.mp3	20110602.WMA 20110602.mp3

Books

There is no textbook for the course. Josh Akey, in last year's web pages, lists some books and a number of on-line statistics texts available free on the web. They are

In fact, a whole bunch of on-line statistics textbooks will be found if you Google: "online statistics text"

Josh's 2008 course web pages are excellent, especially his lecture PDFs. Although the order of material is different, they are very much work looking at. They are here

The R language

R is a free interactive computer environment (in old-fashioned terms, an "interpreter") that can be used for many purposes. It was originally designed by statisticians (R is a clone of a language called S, which is now commercial). It has many built-in statistics functions, which is why we will use it. (At the main CRAN-R project site there are links to many other analysis packages that can be loaded into R).

R can be downloaded and installed on Windows, Mac OS X, or Linux machines (and some other types as well). It is available at the CRAN-R site here as executables, source code, and many other resources including a terse PDF introductory manual. When using this manual skip over parts that go too deeply into stuff you don't yet understand as there is valuable stuff after that. Come back to the skipped stuff later.

Here are some R resources that you may find helpful:

Here is another good set of pages that introduces R.
Christopher Green, of our Statistics Department, produced a good primer on the use of R, aimed at statistics applications. It can be downloaded as a PDF here.
Emanuel Paradis has made available on line a good introduction to R R for Beginners which is also available in a French version.
This set of (PDF) slides by Thomas Lumley of our Biostatistics Department gives much useful information about ways to do things in R.
Two good quick example sheets were made up by Josh Akey two years ago. They are
An R tutorial
and An R descriptive statistics tutorial.
They show you the commands to do some relevant things, but do not show you what appears after you type in the commands. I recommend that you try typing in these commands yourself -- it will be quite helpful.
In past years Brooks Miner and Sylvia Yang of the Biology Department have given a workshop on use of R for biologists, toward the end of Spring Quarter. I don't know whether they will do it again this year (I don't think they will as they are busy).

Is R great? For many things, yes. Is it good at everything? I would say that its array operations stand a good chance of driving the puzzled beginner absolutely bonkers, so no. In this it reminds me of a now-fortunately-defunct programming language called APL ("A Programming Language") which could do many things interactively, had fervent evangelists, was mostly about arrays, and drove me absolutely bonkers. I wrote this assessment two years ago. Since then I have learned more R, but I do not see any reason to change what I said.

R in this course (and R exercises to be done in class)

We will do an R exercise in each class session. Students are expected to bring a laptop with R loaded on it (kudos to this year's class for doing this successfully). I will have exercise sheets for each class, and we will try to do them. As I make them I will post them here.

Exercise 1. Get R working, and try commands on this R introduction sheet and if time on this one too.
Exercise 2. We will do the commands in this exercise sheet which you can view on your browser or use the Save As option in its File menu to save a copy. There is a dataset needed which you will find here and should save as a file called "genes1000", using the Save As menu item in your browser's File menu.
Exercise 3. Here is the exercise sheet. Make sure you get the data set of gene expression levels (from the paper by Storey et al.) which can be downloaded here.
Exercise 4. Use this exercise sheet. It has you use R functions that draw samples from distributions, so you don't need to load any real data.
Exercise 5. Use this exercise sheet. The data set cats.txt, which is data by Janis Abkowitz and Stan Gartler of our own department, is available here in folder "datasets". You will also simulate some data of your own.
Exercise 6. Use this exercise sheet. There are three data sets provided (one other you will simulate yourself):
example6.txt, mussels.txt, and mpi.txt.
Exercise 7. Use this exercise sheet.
Exercise 8. Use this exercise sheet. The data set used is genes1000 which we used before.
Exercise 9. Use this exercise sheet. The data set used is RMA_Filtered.txt which we have used before.
Exercise 10. Use this exercise sheet. Use this data set called "journals.txt".

Homework assignments

Homework assignments will be posted here on Thursdays, to be turned in by email to me by the following Thursday:

Homework 1 is available here. It is to be turned in by the following Thursday, May 12.
Homework 2 is available here. It is to be turned in by the following Thursday, May 19.
Homework 3 is available here. Because it involves material from the May 24 lecture, it does not need to be turned in until Thursday, June 2. The contingency table of data on sick people is available here.
Homework 4 is available here. The 100-gene data set is available here: here. Due Saturday, June 4.

The lecturer

Who is this guy who is teaching the course, anyway?

this page maintained fitfully by Joe Felsenstein