This study, conducted by Viganò et al., attempts to develop a multi-task model that utilizes Artificial Intelligence and Machine Learning methods to predict cardiotoxicity based on prompted small molecules. Cardiovascular diseases are the leading cause of global mortality and are induced by varying risk factors and habits influenced by lifestyle or environmental conditions. The authors built a Mixture of Experts multi-task model that used multiple molecular encodings, such as molecular descriptors and CDDD embeddings, to simultaneously predict 12 different toxicity endpoints. The data that was used to determine the chemicals to run through the model were collected from the NIH ICE database, ChEMBL, and documents related to the FDA. They then compared their multi-task MoE model to a single-task model and a basic multi-task model in a series of measured aspects. Results showed that the single-task model performed poorly on endpoints with few positive samples, and was outperformed by the multi-task MoE in the measured aspects, as the balanced accuracy was 0.74 compared to the single-task model's balanced accuracy of 0.69. Furthermore, they found that the best approaches for encoding chemical information were classical molecular descriptors (MDs) and CDDD. Overall, the MoE multi-task model had better generalization and higher accuracy than traditional single-task models. The MoE-based multi-task model simplifies the evaluation process by using a single model for the prediction of multiple endpoints. Furthermore, this model’s assessment of the cardiotoxic potential of small molecules is more efficient, has improved usability, reduces integration burden, and supports real-world screening and regulatory contexts.