|
Abstract:
|
We consider a situation in which we are confronted with many unlabeled measurements (in R^n) which are generated, with some loss and distortion, from a larger set of possible measurements. The process is repeated across a number of samples, each with its own loss and distortion. We allow the possibility of sample variation in one dimension, but with important correlational structure. Our task is to identify which measurements are repeated across multiple samples and describe the structure in the dimension containing sample variation. We will focus our discussion on the example of high resolution proteomics. With it, one can identify the masses, charge states, hydrophobicities and expression levels of tens of thousands of individual polypeptides concurrently from a single sample. However, these are largely unmarked, therefore matching observations between samples is a challenge.
|
