Statistical methods for Expression Analysis

From: Daniel Chipchase (D.Chipchase@oxfordbiomedica.co.uk)
Date: Thu Jan 17 2002 - 11:23:50 EST


Dear Flowers,
I have been looking at promoter function in transduced cells using a GFP
reporter for some time and am generally unhappy with the rigour of the
'answers' I am getting. What I really require is a statistical method which
gives me a reproducible constant.

The problem is  that the 'true' level of expression should be the
fluorescence I would get off of 1 copy of integrated genome (inc promoter
under investigation and GFP cassette) per cell. However, the number of
copies per cell will vary (normally) I suppose, and also as a result of my
transduction efficiency. As my transduction efficiency improves I would
suppose hypothetically that the modal number of copies per cell would
increase.

Anyway at present I am assuming that at a low transduction percentage any
cells transduced will likely only include (and express) 1 copy of GFP.
Therefore, after removing non-transduced cells, the geometric mean of those
remaining should give me the expression level of GFP, which when normalised
against a control vector gives me a benchmark. However, having reviewed my
accumulated data it is often difficult to see any constant expression at
lower levels of transduction.

Also the means by which I remove the transduced from the non-tranduced cells
is problematic. At low transduction levels where expression may be fairly
low it is often the case that the transduced population is merely a
'shoulder' on the the untransduced population when viewed on a 1 parameter
histogram. I have found no 'satisfactory' way of determining transduction
percentages in these cases.

It has occurred to me that at at transduction levels approaching 100% the
geometic mean might give you an answer that is constant and comparable
between vectors. Might this not be a better approach? (it would use up a lot
of raw materials though).

Hoping someone out there understands what I talking about, and can give me a
few hints

Daniel Chipchase
Research Assistant



This archive was generated by hypermail 2b29 : Sun Jan 05 2003 - 19:25:56 EST