Abstract
Predicting perceived odor intensity remains a fundamental challenge due to the complex, non-linear behavior of olfactory response and the difficulty of correlating molecular structure with human perception. This study introduces VIANA (character Value-enhanced Intensity Assessment via domain-informed Neural Architecture), a novel ''tri-pillar'' framework that serves as the core innovation by integrating structural graph theory, character value embeddings, and phenomenological behavior. This methodology systematically evaluates knowledge transfer across three distinct domains: molecular structure via Graph Convolutional Networks (GCNs), semantic odor character values via Principal Odor Map (POM) embeddings, and biological dose-response logic via Hill’s law. A second major innovation is the implementation of PCA-driven ''signal distillation''; while raw semantic data led to information overload, applying Principal Component Analysis (PCA) to distill the 95% most impactful semantic variance optimized the model’s high-dimensional input space. Results demonstrate that this synthesis significantly outperforms baseline structural models, with VIANA achieving a peak R2 of 0.996 and a test Mean Squared Error (MSE) of 0.19. By bridging molecular informatics and sensory perception, VIANA successfully captures the physical ceiling of saturation and detection thresholds, providing a robust, domain-grounded simulation of the human olfactory experience.