Prediction of double gene knockout measurements

One way to get an insight into the potential interaction between a pair of genes is to compare the phenotype of the double knockout organism to the phenotype we would expect if the genes were independent. However, since gene knockouts are expensive and time-consuming, we would like to be able to predict gene interactions. We used a dataset consisting of 352 yeast genes for which only a few double knockout measurements were known, and applied regression algorithms to predict the unknown measurements. In this article, we describe the algorithms we used and compare their results.