Dense Sense - a novel approach utilizing electron density augmented machine learning paradigm to understand the complex odour landscape

Aug 4, 2026

Olfaction is a complex biological process in which multiple nasal receptors work together to detect odorant molecules, yet structure–activity relationships remain hard to decipher because crystallizing odor receptors is extremely challenging. This study explores a synergy between quantum-mechanics information and graph neural networks to improve prediction of odor perception. The results highlight the feasibility of deriving odor understanding directly from quantum data, providing a novel machine-learning pathway for investigating olfaction mechanisms.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages12

Reading Time30 min

AccessFree to download

Introduction

Challenges in decoding odorant–receptor recognition

Graph neural networks for odor labeling

Ligand-based and quantum-assisted learning approach

Open Access Article . Pu on 08 Octoberblished 2025. Downloaded on 2/2/2026 12:30: 14 PM . Digital Discovery PAPER View Article Online View Journal | View Issue Cite this: Digital Discovery, 2025, 4, 3339 Received 23rd May 2025 Accepted 21st September 2025 DOI: 10.1039/d5dd00224a[rsc.li/digitaldiscovery](rsc.li/digitaldiscovery) Dense Sense: a novel approach utilizing electron density augmented machine learning paradigm to understand the complex odour landscape Pinaki Saha, *a Mrityunjay Sharma,bcd Sarabeshwar Balaji,e Aryan Amit Barsainyan,f Ritesh Kumar, *bc Volker Steuber a and Michael Schmuker a Olfaction is a complex process where multiple nasal receptors interact to detect speciﬁc odorant molecules. Elucidating structure–activity-relationships for odorants and their receptors remains diﬃcult since crystallization of the odor receptors is an extremely diﬃcult process. Therefore, ligand-based approaches that leverage machine learning remain the state of the art for predicting odorant properties for molecules, such as the graph neural network approach used by Lee et al. In this paper we explored how information from quantum mechanics (QM) could synergistically improve the results obtained with the graph neural network. Our ﬁndings underscore the possibility of this methodology in predicting odor perception directly from QM data, oﬀering a novel approach in the machine learning space to understand olfaction. Introduction Olfaction is a vital sense for perceiving the world that is crucial to the survival of many animals, [e.g.](e.g. in)[ in](e.g. in) foraging, mating and detecting prey and predators. It also plays an important role inhuman life, [e.g.](e.g. to)[ to](e.g. to) detect hazards or maintaining hygiene. Although it plays such an important role, the intricacies of the olfaction process are not well understood. In the case of olfaction in humans, olfactory perception involves about 802 genes which encode for the ORs, out of which 388 genes are functional receptors while the remaining 414 are reported to be pseudogenes.5 The exploration of these receptors using structural biology tools has been challenging due to their high genetic variability and limited expression in in vitro systems. Their instability during the isolation process makes them diﬃcult to crystallize.1–3 Additionally, the binding of odorants to the ORs isnot very straightforward. Individual ORs can recognize multiple odorants while a single odorant can elicit responses from multiple receptors, resulting in a complex scheme for odorant recognition.6 The complexity of odorant-OR binding impedes a University of Hertfordshire, UH Biocomputation Group, UK. E-mail: p.saha3@herts. [ac.uk](ac.uk) bCSIR-Central Scienti􀀁c Instruments Organisation, Sector 30-C, Chandigarh-160030, India cAcademy of Scienti􀀁c and Innovative Research (AcSIR), Ghaziabad-201002, India dDepartment of Higher Education, Shimla-171001, Himachal Pradesh, IndiaeIndian Institute of Science Education and Research Bhopal (IISERB), Madhya Pradesh- 462066, India fNational Institute of Technology Karnataka, Surathkal, Karnataka-575025, India structural based understanding of the odorant binding process. Nonetheless, there is some notable research going on concerning the structural biology aspect of olfaction. Utilizing CryoEM, researchers have been able to elucidate the structure of a single human olfactory receptor:7 OR51E2 . In the case of OR51E2 receptor, size selectivity has been observed for carboxylic acid based ligands. Short chain linear carboxylic acids were shown to bind better to OR51E2 receptor compared to their long chain counterparts.3 Recently using the AI based homology modelling tool Alphafold and molecular dynamics (MD) simulation, researchers have tried to elucidate the structure of diﬀerent ORs present in the human nasal epithelium.8 In contrast to the structure based approach, we have the ligand based approach where we solely focus on f

FAQs

Why are traditional structure–activity-relationship studies difficult for odorant receptors?

Crystallization of odor receptors is extremely difficult, making structural understanding hard to obtain.

How does this work improve upon a graph neural network approach for odor prediction?

It combines quantum mechanics (QM) information with graph neural network learning to enhance predictive performance.

What modeling task is the graph neural network used for in odor perception research?

The graph neural network framework predicts odor labels (classification) or related properties through embedding-based learning from molecular graphs.

Aug 4, 2026

Upload by Valentina