Finding Minimal Sets of Informative Genes in Microarray Data

摘要

For a microarray dataset with attached phenotype information -which gives expression levels of various genes and a phenotype classification for each of a set of samples - an important problem is to find informative genes. These genes have high information content as attributes for classification, minimizing the expected number of tests needed to identify a phenotype. This study investigates the use of a heuristic method for finding complete sets of informative genes (sets that are sufficient for constructing a maximally discriminating classifier) that are as small as possible. These minimal sets of informative genes can be very useful in developing an appreciation for the data. Our method uses branch-and-bound depth-first search. Experimental results suggest that our method is effective in finding minimal gene sets, and the resulting classifiers have good performance in terms of classification accuracy.