The minimal number of genes needed to identify a tumor

Abstract

We demonstrate that the global state of a Gene Regulatory Network [1] may be labeled by a few genes in spite of the fact that there are thousands of genes participating in it. For example, the expression values of only 3 genes are enough to discriminate between a tissue sample coming from a normal lung or a lung adenocarcinoma. We follow a pragmatic procedure, dependent on the sample set, but which is expected to become exact for large enough sets of samples. The proof relies on a scheme for the construction of perfect classification panels of genes [2], inspired by rough set theory [3].