To fill this gap, we propose two attention-mechanism-based encoder–decoder models that incorporate multisource information: one is MAEDDI, which can predict DDIs, and the other is MAEDDIE, which can ...