Number of cell lines in training, dev and test dataset for perturbed gene expression profile prediction task and gene expression autoencoder training task.

We evaluated these two tasks with leave new cells out cross-validation. In each split, we leave a group of cell lines out and then split the rest cell lines to training and dev dataset. The cells in training dataset, dev dataset and test dataset are distinct to each other. There are totally 15 cell lines for the perturbed gene expression profile prediction task, which have both gene expression profile found in CCLE and high-quality data in LINCS L1000 project. 11550 cell lines are used for the gene expression autoencoder training and are from both CCLE and TCGA datasets. (XLSX)