Dear DEL-ML challenge participants, A follow-up on Luca's post just below. An important reminder that the active learning phase should not be a simple (or disguised) similarity search. As mentioned earlier, the DEL-ML model should play a significant role in your final selection. For example, using a DEL-ML model to remove the lowest ranking 20% compounds from the test set and using active learning hits to run a similarity search on the remaining 80% does not qualify. We do not want to be overly prescriptive and we rely on your creativity to make sure that important methodological lessons can be drawn once the challenge is completed. Best of luck with the final sprint!