When I participated in the DREAM3 in silico network reconstruction challenge, I had very little experience in reconstructing TF-regulatory networks (I was focusing on protein-protein interaction networks at that time). Given the many sophisticated methods existing at that time, I just hoped to participate and get some experience in this topic. However, perhaps because I was new to this topic, my method was apparently quite different from the other methods, and it seemed to put more emphasis on a new type of features (expression change due to perturbation) than the other methods, which turned out to be very useful. It has helped me to always ask if there are better features to use or better ways to use the existing features before drilling deep into improving subtle details of existing methods. I think this experience has guided my way to dealing with new problems.