VIANA: character Value-enhanced Intensity Assessment via domain-informed Neural Architecture
Authors: Luana P. Queiroz, Icaro S. C. Bernardes, Ana M. Ribeiro, Bernardo M. Aguilera-Mercado, Idelfonso B. R. Nogueira
Organizations: LSRE-LCM, ALiCE, Faculty of Engineering, University of Porto, Rua Dr. Roberto Frias, 4200-465 Porto, Portugal · Department of Chemical Engineering, Norwegian University of Science and Technology, Sem Sælands vei 4, NO-7491 Trondheim, Norway · Corporate Fragrance R&D, The Procter & Gamble Company, Ivorydale Innovation Center, 5299 Spring Grove Ave., Cincinnati, OH 45140, USA · Heritage Science Laboratory, Faculty of Chemistry and Chemical Technology, University of Ljubljana, Večna pot 113, SI-1000 Ljubljana, Slovenia
Predicting the perceived intensity of odorants remains a fundamental challenge in sensory science due to the complex, non-linear behavior of their response, as well as the difficulty in correlating molecular structure with human perception. While traditional deep learning models, such as Graph Convolutional Networks (GCNs), excel at capturing molecular topology, they often fail to account for the biological and perceptual context of olfaction. This study introduces VIANA, a novel "tri-pillar" framework that integrates structural graph theory, character value embeddings, and phenomenological behavior. This methodology systematically evaluates knowledge transfer across three distinct domains: molecular structure via GCNs, semantic odor character values via Principal Odor Map (POM) embeddings, and biological dose-response logic via Hill's law. We demonstrate that knowledge transfer is not inherently positive; rather, a balance must be maintained in the volume of information provided to the model. While raw semantic data led to "information overload" in domain-informed models, applying Principal Component Analysis (PCA) to distill the 95% most impactful semantic variance yielded a superior "signal distillation" effect. Results indicate that the synthesis of these three knowledge transfer pillars significantly outperforms baseline structural models, with VIANA achieving a peak R^2 of 0.996 and a test Mean Squared Error (MSE) of 0.19. In this context, VIANA successfully captures the physical ceiling of saturation, the sensitivity of detection thresholds, and the nuance of odor character value expression, providing a domain grounded simulation of the human olfactory experience. This research provides a robust framework for digital olfaction, effectively bridging the gap between molecular informatics and sensory perception.
Figures & tables
Figure 1: Methodology scheme.
Figure 2: Scheme of the molecular intensity prediction model.
Figure 3: Scheme of the domain-informed molecular intensity prediction model.
Feature
GCN Model
Domain-Informed Model
Output Type
Direct Intensity Prediction ( y )
Parameters ( Imax,C,D )
Inductive Bias
None
Hill’s law
Architecture
Simple GCN + MLP
GCN + LayerNorm + Sigmoidal Head
Optimization
Basic Optuna Search
Optuna with Median Pruning
Physical Consistency
Relies on Data
Hard-coded by Architecture
Stability
Standard Gradients
Gradient Clipping + LayerNorm
Table 1: Comparison between the “pure” GCN and Domain-Informed architectures.
Figure 4: Scheme of the character value-enhanced GCN model.
Figure 5: Scheme of the character value-enhanced domain-informed model.
Figure 6: Scheme of the reduced dimensionality character value-enhanced GCN model.
Figure 7: Scheme of the VIANA model.
Architecture
Opt. MSE
Learning Rate
GCN Hid.
Drop. Rate
Batch Size
POM Weight
Total Epochs
Best Epoch
Val. Loss
Early Stop
Test MSE
R2
GCN (Baseline)
33.56
9.8e-4
64
0.17
32
N/A
120
79
165.79
Yes
195.44
0.010
Domain-informed model
2.6e-3
6.8e-3
64
0.24
64
N/A
210
162
0.04
Yes
0.46
0.991
Character value-enhanced GCN
26.89
7.6e-4
256
0.12
32
1.73
270
220
18.67
Yes
23.47
0.881
Character value-enhanced domain-informed model
0.2920
1.5e-3
128
0.14
32
1.26
190
147
0.23
Yes
0.55
0.989
Reduced dim. char. value-enhanced GCN
18.11
6.7e-4
256
0.14
32
1.80
470
425
10.05
Yes
33.46
0.830
VIANA
0.20
5.2e-4
64
0.11
32
0.72
170
123
0.08
Yes
0.19
0.996
Table 2: Summary of hyperparameter optimization and training performance across all architectures.
Figure 8: Training and Validation MSE Loss per epoch for the “pure” GCN model.
Figure 9: Test set results for the “pure” GCN model.
Figure 10: Residual error results for the “pure” GCN model training.
Figure 11: Test set results for the domain-informed model training.
Figure 12: Residual error results for the domain-informed model training.
Figure 13: Error distribution per intensity range for the domain-informed model training.
Figure 14: Test set results for the character value-enhanced GCN training.
Figure 15: Training and Validation MSE Loss per epoch for the character value-enhanced GCN training.
Figure 16: Error distribution per intensity range for the character value-enhanced GCN model training.
Figure 17: Test set results for the character value-enhanced domain-informed model training.
Figure 18: Training and Validation MSE Loss per epoch for the character value-enhanced domain-informed model training.
Figure 19: Variance analysis on principal components.
Figure 20: Test set results for the reduced dimensionality character value-enhanced GCN training.
Figure 21: Error distribution per intensity range for the reduced dimensionality character value-enhanced GCN training.
Figure 22: Test set results for the VIANA model training.
Figure 23: Training and Validation MSE Loss per epoch for the VIANA model training.
Figure 24: Error distribution per intensity range for the VIANA model training.
Figure 25: Residual error results for the VIANA model training.
Name
True Int.
Baseline
Char. value-enhanced
Reduced dim.
Octyl acetate
14.25
11.58
22.28
29.73
Beta-pinene
16.56
10.49
7.61
8.19
Dihydro-beta-ionone
1.66
10.85
8.92
3.81
Sodium tetrahydrojasmonate
14.69
11.28
18.61
10.19
Coumarin
16.29
11.51
14.88
9.71
Table 3: Comparison of GCN-based architectures on representative test molecules.
Name
True Int.
Domain-informed
Char. value-enhanced
VIANA
p-Tolualdehyde
15.19
14.99
15.85
15.10
Stearyl alcohol
1.30
1.17
1.06
1.34
Isolongifolone
1.36
1.22
1.28
1.49
Alpha-Terpineol
0.77
0.57
0.98
0.83
beta-Santalol
3.62
3.08
5.50
3.55
3,5,5-Trimethylhexan-1-ol
1.88
1.76
1.43
1.89
Table 4: Comparison of Domain-informed architectures leading to VIANA.
Predicting olfactory qualities from molecular structure is an open problem in chemoinformatics. Although linear models can link molecular features to odor descriptors, they often fail when extrapolating to novel chemical scaffolds, extreme molecular weights, or complex odor mixtures. To address this, we introduce GraphNOSE, an open-source graph transformer framework that predicts multi-label odor descriptors from simplified molecular-input line-entry system (SMILES) strings for single molecules and binary mixtures. By integrating positional and structural encodings within a transformer-based graph architecture, GraphNOSE achieves strong performance with six times fewer parameters than standard graph neural network (GNN) baseline while consistently outperforming linear models, molecular language model embeddings, molecular fingerprints, and baseline GNNs by an average area under the ROC curve (AUROC) margin of 4.52% (p < 0.01). GraphNOSE achieves an AUROC of 84% on out-of-distribution compounds (OODs). This exceeds the current state-of-the-art GNN for OOD in olfaction (Open-POM: 81%, p < 0.001), and identifies conditions under which linear models empirically fail. Finally, we apply XAI (explainable AI) methods to identify which substructures and molecular features drive odor predictions, yielding insights consistent with chemical intuition and grounded in the model's learned representations. Together, these results establish GraphNOSE as a scalable and interpretable architecture for olfactory prediction that generalizes to structurally distinct compounds underrepresented in current perceptual databases.
CSIR- Central Scientific Instruments Organisation, Sector 30-C, Chandigarh-160030, India · Academy of Scientific and Innovative Research (AcSIR), Ghaziabad-201002, India · Department of Higher Education,Himachal Pradesh,Shimla -171001,India +3
Olfaction lies at the intersection of chemical structure, neural encoding, and linguistic perception, yet existing representation methods fail to fully capture this pathway. Current approaches typically model only isolated segments of the olfactory pathway, overlooking the complete chain from molecule to receptors to linguistic descriptions. Such fragmentation yields learned embeddings that lack both biological grounding and semantic interpretability. We propose NOSE (Neural Olfactory-Semantic Embedding), a representation learning framework that aligns three modalities along the olfactory pathway: molecular structure, receptor sequence, and natural language description. Rather than simply fusing these signals, we decouple their contributions via orthogonal constraints, preserving the unique encoded information of each modality. To address the sparsity of olfactory language, we introduce a weak positive sample strategy to calibrate semantic similarity, preventing erroneous repulsion of similar odors in the feature space. Extensive experiments demonstrate that NOSE achieves state-of-the-art (SOTA) performance and excellent zero-shot generalization, confirming the strong alignment between its representation space and human olfactory intuition.Code and data are available at https://github.com/Xianyusyy/NOSE
Yanyi Su, Hongshuai Wang, Zhifeng Gao +1
State Key Laboratory of Physical Chemistry of Solid Surface, College of Chemistry and Chemical Engineering, Xiamen University, Xiamen, China · DP Technology · Laboratory of AI for Electrochemistry (AI4EC), Tan Kah Kee Innovation Laboratory (IKKEM), Xiamen, China +1
Olfaction played an indispensable role throughout human evolution and civilization. Even in the contemporary era of advanced technology, olfaction remains a critical channel for person to conduct danger discrimination, emotional experience, and memory formation. However, most substances in nature exist as multi-molecule mixtures. The complexity of mixture compositions, as well as concentration dependent saturation effects and receptor specific activation thresholds, pose substantial challenges in identifying olfactory characteristics. In this study, we proposed a novel bio inspired deep learning framework for accurate odor perception recognition of mixtures. We robustly constructed neural response curves for molecule-receptor interactions, and developed a fusion strategy that integrates attention-weighted multi-receptor curves with concentration-dependent multi-molecule curves, replicating the competitive activation and synergistic integration of mixture components. Furthermore, by comparing the consistency of response curve patterns, the model can transfer knowledge from the semantically rich space of molecular associations to guide recognition of mixture perception characteristics. Therefore, we established a complete computational pathway from chemical blending, neural encoding, to perceptual formation. Finally, we conducted comprehensive evaluation, and results demonstrated exceptional superiority, achieving an accuracy of 92.2%. Consequently, our work provides a generalizable solution to the long standing mixture perception challenge. More importantly, it can be integrated into embodied cognitive systems to enhance the agents perceptual and interactive capabilities in complex scenarios.
Fei Wang, Xiaoya Xie, Junfei Liu +6
College of Biomedical Engineering & Instrument Science, Zhejiang University, Hangzhou, 310012, China · College of Automation Engineering, Nanjing University of Aeronautics and Astronautics, Nanjing 210016, China