I'll explain the problem but what I am looking for is a few suggested methods to approach this problem. You don't need to know what a microarray but if you are interested look here link text The info below is simplified, not addressed to a micro-biologist (I am not one)
There are about 250,000 probes on an array that each measure a part of a dna sample. An array is specific to an individual(germ, person....) If the array is designed for my dna and we put yours on it some probes will match (we are both human I think) and some will not. If they match the measurement is high(the probe glows) as the dna at a probe matched less it glows less. There is a very high variance at each probe across the same dna. So it is very difficult to know if a individual sample matches (same dna) at the probe level.
If the data was nice we could just do a simple Hypothesis test to determine if the sample probe was the same as the dna the array was made for. Because of the high variance this will not work.
We are looking for a way to consider the measurements at other probes to improve the test at a probe. We kinda know this should work (based on the DNA) but we are not sure how to choose the probes that will best assist for each of the set of 250,000 probes or what statistical method would be best to use the information about the other probes to test an individual probe.
Different example, maybe easier to understand:
A person has 6 characteristics, lets say weight, height, age, income, date of birth (db), education, race. I measure characteristics for myself but my measurement tools are really bad so I make several measurements but I have a very large variance in my measurements. Now I measure another person and want to know if they have a different income than myself. But my measurement tools are so bad almost any other person’s measured income would be within my confidence interval. (I can only measure the other person once, It kills them ☺ ) But I know that other characteristics should be able to help me determine if the income is the same. I don’t know which ones (weight, height, age, db, education, race). I need a method to choose the most useful/helpful characteristics. Now lets say I know that height and age can help determine if our income are the same or different. What would be the statistical method used?
I have what might be called a training set. That is in know what is and is not the same.
@sheldon-cooper answer, "Note: neither of this will give a "significance score"", A confidence value is good enough. I don't need a statistical "Test"
As an MD I hope that this is an acceptable question for MathOverFlow.net. and I hope you can understand the question as abridged as it is. If some would like to help me rewrite or has suggestion on how I can rewrite the question without all the bio background and as a more specific stats question, let me know.