What datasets were used to train the CertisAI pre-trained model initially?

Three different datasets were used to initially train the first CertisAI pre-trained model. Those datasets include: 1) Genomics and drug sensitivity cell lines, 2) A drug combination dataset from drugcombo.org), and 3) Clinical data from TCGA (mostly chemotherapy drugs). In total, the CertisAI pretrained model was initially trained on more than 4,500 drugs across varying cancer indications.