Path: news.mathworks.com!not-for-mail
From: "Roger Stafford" <ellieandrogerxyzzy@mindspring.com.invalid>
Newsgroups: comp.soft-sys.matlab
Subject: Re: Is this kind of regression possible?
Date: Sat, 1 Dec 2007 04:12:44 +0000 (UTC)
Organization: The MathWorks, Inc.
Lines: 55
Message-ID: <fiqmvs$c6o$1@fred.mathworks.com>
References: <fio8nm$6k4$1@fred.mathworks.com> <fiqisu$l62$1@fred.mathworks.com> <fiqkd0$a9h$1@fred.mathworks.com>
Reply-To: "Roger Stafford" <ellieandrogerxyzzy@mindspring.com.invalid>
NNTP-Posting-Host: webapp-05-blr.mathworks.com
Content-Type: text/plain; charset="ISO-8859-1"
Content-Transfer-Encoding: 8bit
X-Trace: fred.mathworks.com 1196482364 12504 172.30.248.35 (1 Dec 2007 04:12:44 GMT)
X-Complaints-To: news@mathworks.com
NNTP-Posting-Date: Sat, 1 Dec 2007 04:12:44 +0000 (UTC)
X-Newsreader: MATLAB Central Newsreader 1187260
Xref: news.mathworks.com comp.soft-sys.matlab:440319


"vicky " <vivek_mutalik@yahoo.com> wrote in message <fiqkd0$a9h
$1@fred.mathworks.com>...
> Hi Roger,
> 
> Thanks very much for responding to my query and offering to
> help me out. 
> 
> I must say, You have understood my question correctly. I
> appreciate your solution to the problem. Ive tried that.
> taken just binary 1 or 0 representation for each base
> instead of 1000, 0100, type. Im thankful to Walter Roberson
> who gave me that routine.
> 
> Most important issue with my result was that my matrix X,
> which describes these binary representations is not full
> rank. I think some columns are "collinear (or
> multicollinear)". Matrix Y is my activity vector. So i was
> trying to use PLSR to avoid this collinearity and to include
> full DNA sequence (about 50 letter length), which was also
> not successfull due to lack of correlations. Im still
> working on that.
> 
> Meanwhile, I was thinking of why not just take all equations
> in one residual sum of square equation and solve it for
> smaller segments of DNA (Not sure if optimization toolbox
> would be helpful in that). I was trying to eliminate 'T' so
> that i can have less variables to handle. 
> Someone suggested that this may be easy in mathematica. I
> dont understand how will that be. If it is easy, then Matlab
> will be my choice.
> Thanks again. Do send your comments.
> Regards,
> Vivek
--------
  First of all, my apologies for the multiplicity of copies of my previous reply; I 
think there were seven in all, much to my disgust.  I waited for several 
minutes to elicit a response from the MathWorks' newsreader and then 
clicked the "post message" button again.  After many minutes I clicked once 
more, but I can't imagine where seven copies came from &#8211; unless possibly my 
mouse click bounced repeatedly.

  Back to the subject at hand, when you said "taken just binary 1 or 0 
representation for each base instead of 1000, 0100, type" you gave me the 
impression that you interpreted the sequences like 1 0 0 0 or 0 1 0 0 as 
being single binary scalars.  That isn't what I meant.  These are to be four 
distinct elements with each element a 0 or 1.  This makes M have 4*n 
columns and W have 4*n rows.

  I don't understand what you mean in your final paragraph, "Meanwhile, I was 
thinking of why not just take all equations in one residual sum of square 
equation and solve it for smaller segments of DNA".  I see no particular 
reason for taking smaller segments.  What is there to gain by that?

Roger Stafford