Deep-learning model boosts accuracy in predicting ischemic stroke risk with atrial fibrillation
Deep-learning model boosts accuracy in predicting ischemic stroke risk with atrial fibrillation lead image
Atrial fibrillation poses a major risk of ischemic stroke to patients, with multiple classes of medications aimed at reducing this risk. Current risk assessment models, however, struggle to draw together the myriad complex interactions between drugs and proteins in the disease process.
Researchers have developed a network-based deep learning model that combines drug-protein-disease pathways and real-world clinical data to predict one-year risk for ischemic stroke in patients with atrial fibrillation. Using a heterogeneous multilayer network and data from 7859 patients with atrial fibrillation, the model by Lyu et al., called AF-Biological-IS-Path (ABioSPATH), combines mechanistic pathways with patient-specific details to provide individualized risk assessments based on drug mechanisms and how comorbid diseases propagate.
“Unlike traditional approaches that analyze comorbidities and medications in isolation, our model connects drug-protein interactions, disease-protein relationships, protein-protein interactions, and disease-disease associations into a unified framework,” said author Qingpeng Zhang. “This integration allows simultaneous exploration of molecular mechanisms underlying stroke development while improving prediction accuracy.”
Compared to baseline prediction models, ABioSPATH delivered superior performance, with individual-level analysis demonstrating the importance of the PIK3/Akt pathway, a key regulator in how the cell cycle controls cellular proliferation and quiescence.
It also showed stroke risk associated with drugs less associated with stroke risk, such as prochlorperazine, a medication commonly used for nausea, migraines and psychosis rarely discussed clinically in stroke mitigation.
Cohort-level analysis highlighted the importance of easily obtainable lab values such as the nonspecific inflammatory marker C-reactive protein, or kidney-produced blood pressure regulatory protein renin, as well as less clinically used markers such as prostaglandin synthases.
The group hopes the model pushes forward both clinical risk stratification of stroke and the use of AI for disease prediction.
Source: “Predicting the risk of ischemic stroke in patients with atrial fibrillation using heterogeneous drug-protein-disease network-based deep learning,” by Zhiheng Lyu, Jiannan Yang, Zhongzhi Xu, Weilan Wang, Weibin Cheng, Kwok-Leung Tsui, and Qingpeng Zhang, APL Bioengineering (2025). The article can be accessed at https://doi.org/10.1063/5.0242570