DEEP LEARNING FOR MOLECULAR PROPERTY PREDICTION OF NOVEL COMPOUND

Abstract

Images of Molecules are broadly used to foresee atom properties in educating and synthetic exploration. A prepared physicist can undoubtedly infer particle properties by dissecting its construction and assess its useful gatherings. To foresee, for instance, the water dissolvability of a natural compound a scientist would instinctively tally the quantity of polar gatherings, think about the size of the molecule, and gauge the water/atom communication by tallying the quantity of H-security benefactors and acceptors. Subsequently, 2D molecule portrayals and their straightforwardly available highlights ought to give sufficient data to anticipate the atom's design subordinate properties. To help this postulation, image based AI approach convolutional neural network, clustering and transfer learning are utilized and assessed in this work. At long last properties of new molecule produced utilizing deep learning were anticipated.