Bioactivity prediction using convolutional neural network

Abstract

According to the similar property principle, structurally similar compounds exhibit very similar properties as well as similar biological activities. Many researchers have applied this principle to discover novel drugs, thereby leading to the emergence of the prediction of the activities of compounds based on their chemical structure, since the toxic or biological properties of compounds are determined by their chemical structure, particularly, their substructures. The concept of functional groups (FGs) of connected atoms (small molecules) determining the properties and reactivity of the parent molecule forms the cornerstone of organic chemistry, medicinal chemistry, toxicity assessments and QSAR. This study introduced a novel predictive system, i.e., a convolutional neural network that enables the prediction of molecular bioac-tivities using a novel molecular matrix representation. The number of atoms in small molecules were investigated to determine its accuracy during the prediction of the activities of the orphan compounds. This approach was applied to popular datasets and the performance of this system was compared with three other classical ML algorithms. All the experiments indicated that the proposed model was able to provide an interesting prediction rate (accuracy of 90.21).