Created by Matthias Lienhard mats2704 The delay in launching Round 2 was very unfortunate but we could not anticipate it when starting this challenge. On the bright side, it actually provided the participants with much more time to work on their models. In particular, re-opened Round 1, i.e., Round 1b, was specifically designed to facilitate model improvements. This round closed only a week ago. Please note that Round 2 asks exactly the same question as Round 1, just using different compound-kinase pairs to evaluate the predictions. The purpose of Round 2 is to further assess the generalisability of the models as well as to select the winners of the challenge (winning metrics are revealed now). Thus, if a model performed well in Round 1a and/or 1b, it is likely that it will perform well in Round 2. Even though you might of course still use this Round for further model improvements, in principle, training new models should not be necessary. Anna:
Is there a particular need to make such a short deadline? First, you gave us delay for many months and postponed deadlines indefinitely. Next, you reverse this decision and put a strict timeline.
This is not a very consistent policy from organizers. In addition to frustrating experience, this is now very demanding on our efforts. Plus some of these models take really long time to train. Round 2 will close on April 18, 2019.