MARSY (Multitask drug pAiR SynergY) is a deep learning-based multitask model developed for predicting drug combination synergy scores in cancer. It incorporates information on the gene expression profiles of cancer cell lines and the differential expression signatures induced by individual drugs. MARSY aims to address the sparsity of large drug screening databases by accurately imputing missing values. The model utilizes two encoders to capture the interplay between drug pairs and drug pairs with cell lines, learning latent embeddings that enhance prediction performance compared to state-of-the-art and traditional machine learning models. It has been used to predict synergy scores for over 133,000 new drug-pair cell line combinations, offering insights for precision medicine and drug development.