Fig 1. Overview of the DREAM9 AML-OPC data, implementation, and timeline.
Samples were collected from 291 patients diagnosed with AML and the levels of 231 signaling proteins were assayed using RPPA. Data was selected from 191 of these patients and released to participants on June 16th 2014 for model training. DREAM9 AML-OPC consisted of 3 subchallenges (SC1, SC2 and SC3) which were evaluated independently on a set of 100 patients (the evaluation was performed on different subsets of 70 patients every week), asking participants to predict either Response to Therapy (RT), Remission Duration (RD), or Overall Survival Time (OS). Participants were given weekly feedback on model performance, which was evaluated using two different metrics for each subchallenge, until September 15th 2014 when the challenge concluded. The metrics were Area Under the Receiver Operating Characteristic (AUROC) curve and Balanced Accuracy (BAC) for SC1, and Concordance Index (CI) and Pearson Correlation (PC) for SC2 and SC3. A hackathon was organized during the Challenge to foster collaboration between participants