Therefore, I construct a co-expression network using a k-nearest-neighbor method, where each gene is connected to k other genes with whom its expression profile is most similar. The expression of a prediction target under a prediction condition is then estimated to be the average of the expression levels of its k nearest neighbors, under the same condition. Interestingly, this idea coincides with one of the simplest missing data imputation methods. Indeed, the challenge problem is exactly an example of a missing value estimation problem, for which many algorithms have been developed. This simple method turns out to work well. Among the nine methods that made the final predictions, it shares the ����best performer���� honor with a much sophisticated method, which is based on soft integration of multiple data types using elastic net. The performance of the two top-ranked algorithms is almost identical, and is much better than that of the other participating methods. In addition, I also proposed several alternative strategies, all based on simple ideas for missing value imputation. These 3-Dehydroverticine results were not submitted to the challenge organizers officially. In particular, a modified KNN method achieved even better accuracy than the standard KNN method. Another KNN-based approach did not improve over the standard KNN, while a regression-based approach had slightly lower accuracy than the KNN-based methods. These results, together with the fact that none of the top-performing methods are trying to explicitly construct gene regulatory networks seem to confirm my hypothesis that (-)-Licarin-B current gene regulatory models are probably not accurate enough to model gene expression yet. In addition, the results also suggest that simple methods should in general be preferred over complex ones. The remainder of this paper is organized as follows. In the next section, I first present the challenge problem, and then describe the prediction results I submitted to DREAM3. I also present the results from several alternative strategies and discuss the difference between several evaluation methods for measuring prediction accuracy. I then discuss some lessons learned from my participation in this challenge.