Artificial Neural Network prediction of specific VOCs and blended VOCs for various concentrations from the olfactory receptor firing rates of Drosophila melanogaster

In our previous work, we have investigated the classification of odorants based on their chemical classes only, e.g. Alcohol, Terpene or Ester, using Artificial Neural Networks (ANN) as the signal processing backend of an insect olfactory electronic nose, or e-nose. However, potential applications of e-noses in the food and beverage industry which include the assessment of a fruit's ripeness, quality of wines or identifying bacterial contamination in products, demand the ability to predict beyond chemical class and to identify exact chemicals, known as specific Volatile Organic Compounds (VOCs) and blends of chemical that present themselves as aromas, known as blended VOCs (BVOCs). In this work, we demonstrate for the first time how it is possible to predict such VOCs and also BVOCs at varying concentration levels. We achieve this goal by using ANNs in the form of hybrid Multi-Layer Perceptrons (MLPs), to analyze the firing rate responses of the model organism Drosophila melanogaster's odorant receptors (DmOrs), in order to predict the specific VOCs and BVOCs. We report for the raw and noise injected data how the highest MLP prediction for specific VOCs occurred at a 10-4mol.dm-3 concentration in which all the VOC validation vectors were identified and at a concentration of 10-2mol.dm-3 for BVOCs in which 8/9 or 88.9% were identified. The results demonstrate for the first time the power of using MLPs and insect odorant receptors (Ors) to predict specific VOCs and BVOCs.