DDIPrompt: Drug-Drug Interaction Event Prediction based on Graph Prompt Learning
CoRR(2024)
摘要
Recently, Graph Neural Networks have become increasingly prevalent in
predicting adverse drug-drug interactions (DDI) due to their proficiency in
modeling the intricate associations between atoms and functional groups within
and across drug molecules. However, they are still hindered by two significant
challenges: (1) the issue of highly imbalanced event distribution, which is a
common but critical problem in medical datasets where certain interactions are
vastly underrepresented. This imbalance poses a substantial barrier to
achieving accurate and reliable DDI predictions. (2) the scarcity of labeled
data for rare events, which is a pervasive issue in the medical field where
rare yet potentially critical interactions are often overlooked or
under-studied due to limited available data. In response, we offer DDIPrompt,
an innovative panacea inspired by the recent advancements in graph prompting.
Our framework aims to address these issues by leveraging the intrinsic
knowledge from pre-trained models, which can be efficiently deployed with
minimal downstream data. Specifically, to solve the first challenge, DDIPrompt
employs augmented links between drugs, considering both structural and
interactive proximity. It features a hierarchical pre-training strategy that
comprehends intra-molecular structures and inter-molecular interactions,
fostering a comprehensive and unbiased understanding of drug properties. For
the second challenge, we implement a prototype-enhanced prompting mechanism
during inference. This mechanism, refined by few-shot examples from each
category, effectively harnesses the rich pre-training knowledge to enhance
prediction accuracy, particularly for these rare but crucial interactions.
Comprehensive evaluations on two benchmark datasets demonstrate the superiority
of DDIPrompt, particularly in predicting rare DDI events.
更多查看译文
AI 理解论文
溯源树
样例
生成溯源树,研究论文发展脉络
Chat Paper
正在生成论文摘要