One of my favorite in-class exercises is The Assembly Exercise, in which I provided "shotgun sequence" from some English text and ask the students to assemble it. Normally I provide a printout of about 10-20 pages of reads with range of read lengths, error rates, and single/paired end sequences, all from the same sample. My experience has been that biology students get a much better visceral sense of the problems of de novo assembly this way than through any amount of lectures and problem sets.
Until today, I had always used a crappy set of ad hoc scripts to generate the data. I've just put these scripts into a CGI-based Web form, available here:
http://lyorn.idyll.org/~t/assembly-exercise/index.cgi
and you can get the source code here,
https://github.com/ged-lab/assembly-exercise
on github.
The scripts let you put in any text & choose your mutation rate, read length, and desired coverage. You can also set whether you want paired ends, along with the insert size you want.
So, you want to try this out, eh?
Making this a class exercise
Setup
First, go find a paragraph of text, and then find some sentences that overlap with that paragraph in some way (so you can get some repeat sequences in there).
Then, generate ~8-15 base reads with a 1-2% error rate to a combined coverage of about 5. Depending on what you want to highlight (the role of error? the frustration of short reads?) you can generate another set or two with different parameters.
Finally, generate some paired end reads; I recommend a read length of about 5, together with an insert size of 15-25.
Print all of these out.
Instructions
Print out enough copies so that you can divide your class up into groups of 3-5 and give each group their own set of copies.
Tell the students:
- Here are a bunch of subsequences, sampled with error, from some text.
- These are all from the same text, but they may have different read lengths and error rates.
- The word,word output (from paired ends) represent the ends of a larger fragment from the text. That is, you have the additional information that each pair of words is some specific distance apart.
- Please use pen/paper or a computer notepad to figure out what the source sentences are.
- No, you may not use Google.
Also ask them to figure out what a generalizable strategy might be, and how they can verify that the sequences they get at the end are correct.
I don't provide scissors and glue, although it'd be interesting to see what would happen if I did -- sticky graduate students? :)
Duration
I recommend about 45+ minutes of work in the groups, plus 5 minutes intro and 10-20 minutes of discussion at the end. It fits well into an 80 minute class.
Discussion
I usually try to touch on the following points.
Data presentation and information:
- sorting doesn't actually help all that much.
- knowing what the coverage is can be helpful, since it lets you detect repetitive regions that will lead to misassemblies.
- longer reads are way more helpful than shorter :)
Algorithms:
- a word-based algorithm, where you group sentences by shared (long) words are great, IF you have low error data. (This is the de Bruijn graph approach.)
- many students use a greedy algorithm, where they try to find the first possible extension of a sentence fragment. This leads to misassemblies!
- no student has ever used an alignment based approach -- it's too slow for humans!
- if you did one pass where you corrected all the errors, your life would be way better!
Validation:
- the results should be English (or whatever language you input). That's really a huge advantage.
- the paired ends are great for validation, if you didn't use them in the assembly process.
Follow-up reading
I highly recommend assigning "Assembly algorithms for next-generation sequencing data" (JR Miller et al.), http://www.ncbi.nlm.nih.gov/pubmed/20211242, for follow-up reading.
---
Enjoy!
--titus
Comments !