Assessing the performance of the multitask model. (A) Scatter plots comparing the AUROC and AP scores of DeepRiPe and the singletask models. Single models 1 and single models 2 are trained on random negative samples from binding sites of other RBPs and unbound transcript, respectively. Each data point represents an RBP and it falls above the diagonal when DeepRiPe outperforms its singletask counterpart. (B) Comparing the attribution maps obtained from the multitask and singletask models using the IG method for two positives samples of ELAVL2.