Graph Neural Networks based Model for Aroma Prediction using Molecular Structures

In scientific and forensic research investigations, predicting sensory systems stimulus as evidence is inevitable. It is a decades-old challenge in the field of cheminformatics, which can be solved using various Learning techniques. These techniques have made a good impact in Computer Vision (CV), Nature Language Processing (NLP) and Speech Recognition (SR). Among sensory systems, the Olfaction Mechanism is moderately under-explored due to its high complexity in odorant molecular structures. Advanced computers facilitate the machine learning techniques to classify into clusters and predict the aroma. An attempt is made to embed Graph Neural Networks (GNNs) for olfactory senses prediction, significantly encompasses the traditional methods. In this investigation, the dataset curated by the olfactory experts is used for transfer learning which is based on the spatial odor representations (pattern recognition) of the molecular structures of the substance. Aroma prediction plays a key role in various research domains including synthetic-fragrance manufacturing, human dietetics, sensory neurosciences, and forensic crime investigations.