Abstract
In this paper, we introduce a unified and generalist Biomedical GenerativePre-trained Transformer (BiomedGPT) model, which leverages self-supervision onlarge and diverse datasets to accept multi-modal inputs and perform a range ofdownstream tasks. Our experiments demonstrate that BiomedGPT delivers expansiveand inclusive representations of biomedical data, outperforming the majority ofpreceding state-of-the-art models across five distinct tasks with 20 publicdatasets spanning over 15 unique biomedical modalities. Through the ablationstudy, we also showcase the efficacy of our multi-modal and multi-taskpretraining approach in transferring knowledge to previously unseen data.Overall, our work presents a significant step forward in developing unified andgeneralist models for biomedicine, with far-reaching implications for improvinghealthcare outcomes.