TLDR: ATHENA is a novel hierarchical graph neural network framework that integrates patient clinical features with molecular interaction data to improve the classification of subclinical atherosclerosis. It outperforms existing methods by leveraging both cohort-wide patterns and individual molecular profiles. Crucially, ATHENA also identifies distinct molecular patient subtypes within clinically defined atherosclerosis categories, offering explainable insights for personalized treatment strategies.
Atherosclerosis, a condition where plaque builds up in the arteries, is a silent precursor to serious cardiovascular events. Its insidious nature often means it goes unnoticed until significant damage occurs, making early and accurate detection crucial for effective intervention. Current diagnostic methods, such as imaging, can be inconsistent, and traditional machine learning approaches often overlook the complex, interconnected biological mechanisms at play within a patient’s body.
Addressing these challenges, researchers have introduced a novel framework called ATHENA: Atherosclerosis Through Hierarchical Explainable Neural Network Analysis. This innovative approach leverages a hierarchical graph neural network to integrate two critical types of patient information: broad clinical features that reflect patterns across a group of patients, and unique molecular data specific to an individual’s biological makeup.
Unlike previous methods that might focus solely on molecular fingerprints or clinical similarities in isolation, ATHENA builds a sophisticated, multi-layered network representation. It embeds patient-specific protein-protein interaction (PPI) networks, derived from individual molecular data like transcriptomics, within a larger patient similarity network constructed from clinical features. This dual integration allows ATHENA to capture both the unique molecular mechanisms within a patient and how they relate to broader cohort-wide patterns of the disease.
The power of ATHENA lies in its ability to not only improve the classification of subclinical atherosclerosis but also to provide explainable insights. By using an Explainable AI (XAI) module, ATHENA can identify the specific molecular interactions most influential in a patient’s prediction, essentially pinpointing potential biomarkers. This capability is vital for understanding the underlying disease pathology and guiding personalized treatment strategies.
Evaluated on clinical datasets, including one from the PESA study with 391 patients and another from Steenman et al. with 104 patients, ATHENA demonstrated superior performance. It significantly boosted classification accuracy, showing improvements of up to 13% in AUC and 20% in F1 score compared to various baseline models, including linear methods, deep learning models, and other graph deep learning approaches. This highlights that the synergistic combination of clinical and molecular data within a hierarchical structure is key to its effectiveness.
Beyond classification, ATHENA has made a significant discovery in patient subtyping. While clinicians traditionally categorize atherosclerosis into subtypes like Generalized, Intermediate, and Focal based on imaging, ATHENA revealed two distinct molecularly-driven clusters within each of these clinically defined subtypes. For instance, in generalized atherosclerosis, one cluster was characterized by inflammatory pathways, while another showed signs of structural remodeling. Similarly, intermediate and focal subtypes also exhibited unique molecular distinctions, suggesting different stages or mechanisms of disease progression even within the same clinical diagnosis.
Also Read:
- Advancing Drug Discovery: A New AI Model for Precise Compound-Protein Affinity Prediction
- Enhancing Graph Learning with External Knowledge and Latent Space Constraints
These findings underscore the complexity of atherosclerosis and the potential for more precise, personalized interventions. By understanding these molecularly informed patient clusters, healthcare providers could potentially tailor treatments more effectively, leading to improved patient outcomes. While further longitudinal studies and the integration of additional ‘omics’ data layers are future considerations, ATHENA represents a significant step forward in applying advanced AI to unravel the intricacies of cardiovascular diseases. For more detailed information, you can refer to the full research paper: Atherosclerosis through Hierarchical Explainable Neural Network Analysis.
