Re: [MLUG - DISCUSSION] these are the days of miracles and wonders: a radically new sequencing machine appears

[Jonathan King <EMAIL:PROTECTED>]

On 8/1/05, Stephen Montgomery-Smith <EMAIL:PROTECTED> wrote:
> Jonathan King wrote:
> > On 8/1/05, Stephen Montgomery-Smith <EMAIL:PROTECTED> wrote:

> >>I picked up a science magazine in Tucker Hall the other day.  On its
> >>cover was the lead in to one of its articles "4 ways to produce better
> >>microarrays" (just like "10 ways to get your man to love you more" that
> >>you find in grocery store check out line magazines).
> 
> > Yeah, sad to say, biologists do fall for this kind of thing. :-)
> 
> Of course computer magazines are no different ("15 ways to avoid getting
> your computer hacked").

No, it's all the same.  All trade magazines are pretty much identical
in form.  You'd like to think the content is different, of course.  I
guess mathematicians don't really have these, since they don't need
much gear and probably wouldn't buy much if they did. :-)
 
> > This DNA is then what you use (after a bit more processing I'll skip)
> > on the microarray AKA gene chip AKA DNA chip.
> 
> So I've read a bit more of the Watson book, so I can take a bit more
> detail.  Do you make the DNA into a single strand?  Is the DNA still one
> long piece, or has it been chopped into small pieces?

I don't think you have to chop it up; you do need to make it
fluorescent before you spot it on the chip.  A surprisingly good
article on this is:

http://en.wikipedia.org/wiki/DNA_microarray

I know the probes themselves can be quite large, although Affymetrix
has made quite a name for themselves using relatively short probes
(but lots of them).

> >  Basically, you take a
> > minute spot of this stuff and apply it to a specific tiny piece of the
> > chip that has been "printed" with a short stretch of DNA probe (a
> > complementary sequence) that some of your sample might stick to.  With
> > correct treatment and processing, you can then measure the amount of
> > sample that stuck to the probe, and compare it to what stuck to other
> > places on the chip that have different probes.
> 
> I'm trying to get an idea of what kind of statistical problems you might
> run into.  About how many pieces of DNA stick to one piece of probe?

Ah...now you're talking. :-)  And this is one of the potential
problems; how efficiently you hybridize depends on a lot of things. 
How do you best measure the weak fluorescences?  What strength of
evidence do you need to make a "call"?  How do you best use control
probles (essentially "blank spots")?  DNA is surprisingly compact
stuff, but then so are micro-arrays.

> And are long pieces of similar probes next to each other, or are the
> adjacent probe pieces all totally different?

I think people now do try to mix it up.  And replication has become an
important feature of most protocols.

>  How many base pairs long
> is the typical probe, and how specific is it in how well it attaches to
> a piece of DNA (i.e. will it stick to a piece of DNA that gets 19 base
> pairs out of 20 correct)?

Affymetrix uses probes that are like 20-25 bases, and I think they
claim high stringency.  But, yeah, if you're using some PCR product as
your probe and it's hundreds of bp long, there is a question about how
specific your hybridization will have to be.  I'm glad that you've
appreciated that this is subtle stuff.  And even if you had completely
reliable data from the array itself, interpretation can be a
nightmare.  So suppose you had samples from a developing organism at
many different stages of development.  All other things being equal,
you would expect that genes whose expression patterns are highly
similar are at least somewhat likely to participate in similar
processes.  But this is a quintessential correlation is not causality
problem.

Molecular biology is increasingly becoming applied mathematics.  So
another beautiful technique is yeast two hybrid screening:

http://en.wikipedia.org/wiki/Two-hybrid_screening

Now our problem is to find which proteins interact with each other to
make things really happen in cells.  The good news is that we can do a
good job of determining which proteins *can* interact.  This isn't the
same thing as which two proteins *do* interact, of course.  Maybe you
can use microarrays to show that two genes that code for proteins
really do occur in the same cell types...but eventually you have to do
experiments.
 
There haven't been any Nobel Prizes awarded for bioinformatics...yet. 
This may well be the field where mathematicians will have their best
shot at winning a Nobel or three, though.

jking

_______________________________________________
discussion mailing list
EMAIL:PROTECTED
http://mlug.missouri.edu/mailman/listinfo/discussion