Pancreatic cancers are widely recognized as a group of neoplasms with one of the poorest prognoses in oncology research. Despite the advances achieved in drug design and development, there is no effective cure for pancreatic cancers, and the current chemotherapeutic regimens increase the survival rate by only a few months. As an integral part of all modern drug discovery campaigns, computer-aided approaches can represent a promising alternative change to accelerate the early discovery of potent anti-pancreatic cancer agents. To date, however, most of the efforts made so far have focused on small series of structurally related chemicals, where the anti-pancreatic cancer activity has been measured against only one cancer cell line. In addition, no rational insight has been provided in the sense of unveiling the physicochemical aspects and the structural features that the molecules should possess to increase the anti-pancreatic cancer activity. This work reports the first multicellular target QSAR model based on ensemble learning (mct-QSAR-EL) that allows the simultaneous prediction and design of molecules with activity against different pancreatic cancer cell lines, which exhibit different degrees of sensitivity to chemical treatment. The mct-QSAR-EL model displayed sensitivities and specificities higher than 80% in both training and test sets. The physicochemical and structural interpretations of the molecular descriptors in the model permitted the selection of several fragments with potentially positive contributions to the increase of the anti-pancreatic cancer activity. These fragments were then assembled to design new molecules. The designed molecules were predicted as multicell line inhibitors by the mct-QSAR-EL model, and these results converged with the predictions performed by recently reported models. The designed molecules complied with Lipinski's rule of five and its variants.