Hi, I've probably missed something but what type of measurements are we predicting, e.g. IC20s, IC50s, AUCs, . . . ? Is there any training set that we can use? The unconcealed perturbed RNAseq profiles are useful for that but do we have the sensitivity of the cell lines to those compounds? Thanks in advance and thank you for setting up such as nice challenge. Kind regards,

Created by Emanuel Goncalves emanuel
Hi Emanuel, The AUCs are constrained by the maximum dose in each drug. The AUCs for this challenge are taken from CTRP, not from the data contributor, so you should be able to map the dataset with the unconcealed compounds to the CTRP data. Best, Robert
Hi Robert, Great, thanks for the clarifications. Are the AUCs constrained by the maximum dose used in each drug? And do we have the AUCs for the unconcealed compounds? Thank you,
Hi Emanuel, You are predicting a confidence of sensitivity for each cell line, which is a continuous value between and including 0 and 1, where 1 indicates 100% confidence that the cell line is sensitive and 0 indicates 100% confidence that the cell line is not sensitive. From there, we calculate the AUROC based on a binarized assessment of sensitivity for each drug as defined by mixture modeling of the dose response AUC values. We also calculate the Spearman correlation of your confidence scores to the dose response AUC values. So, both scores are based on sensitivity as defined by drug AUC, but the challenge metrics assess two different things - accurate classification and accurate ranking.

Predict drug-sensitivity measurement type page is loading…