We have developed a novel algorithm for analyzing gene expression data. Thi
s algorithm uses fuzzy logic to transform expression values into qualitativ
e descriptors that can be evaluated by using a set of heuristic rules. In o
ur tests we designed a model to find triplets of activators, repressors, an
d targets in a yeast gene expression data set. For the conditions tested, t
he predictions made by the algorithm agree well with experimental data in t
he literature. The algorithm can also assist in determining the function of
uncharacterized proteins and is able to detect a substantially larger numb
er of transcription factors than could be found at random. This technology
extends current techniques such as clustering in that it allows the user to
generate a connected network of genes using only expression data.