Molecular Descriptor to Classify the Different Isomers of Cocaine
Luis Puerta12, Carlos Gonzalez2
1 Departamento de Química, FACYT, Universidad de Carabobo, Apartado 2005, Valencia, Estado Carabobo, República Bolivariana de Venezuela
2 Chemical and Biochemical Reference Data Division, NIST, Gaithersburg, Maryland 20899, USA
Cocaine is an alkaloid ester extracted from leaves of plants including family Erythroxylaceae1 . According to its molecular geometry four diastereomers showing one pair of enantiomers each can be predicted. The biological activity for each isomer has been measured experimentally2 and the enantiomer R(-)cocaine has shown the greatest activity. We have performed electronic structure calculations to obtain the molecular descriptors: dipole moments (µ), LUMO energies (ELUMO), and the specific rotation (a) for each isomer. We then propose the use of these descriptors as inputs to a Artificil Neural Network (ANN) classifier implemented as a multilayer perceptron (MLP), based on back propagation learning algorithm3 . Neurons in the MLP layers were distributed in the following way: five hidden, eight output, and input that could be either two or three depending the case. Firstly, we considered µ and ELUMO as molecular descriptors. It was found that this simple ANN approach was able to classify the different cocaine diastereomers. In addition, when a was also included as a third descriptor, the further classification of each enantiomer was also possible. The relatively small means square error (less than 0.005) suggest that the use of these three molecular descriptors might be suitable to classify a greater number of illicit drugs similar to cocaine.
Reference
1. https://en.wikipedia.org/wiki/Coca
2. Satendra Singh. Chem. Rev. 2000, 100, 925-1024.
3. Neuroph 2.94 (Ševarac 2017)