Prediction of active compounds from SMILES codes using backpropagation algorithm
Dian Eka Ratnawati, Marjono Marjono, Syaiful Anam · AIP conference proceedings · 2018
An active compound is a compound which has therapeutic and pharmacological activity that directly impacts the diagnosis, recuperation, palliation, therapy, or prevention of disease. Every active compound has a Simplified Molecular Input Line Entry System (SMILES) code. SMILES, proposed by Weininger, is a chemical notation system which represents a simple molecule structure. There are correlations between a molecule’s structure and its activities in that chemicals of similar structures have similar biological activities. This paper will find patterns of SMILES code using a backpropagation algorithm. This method is chosen because of its good ability to make predictions. Prediction of compound activities is interesting because there are a large number of active compounds in the PubChem dataset, but only a few have a known activity. To find an active compound function one uses laboratory tests to extract the compound’s activities. To predict the compound’s activities, the first phase is to extracts features of the SMILES code. After that, these features are extracted are used as inputs to backpropagation algorithms. The results of testing show that these backpropagation algorithms are successful at predicting an active compound, with an average accuracy of 87.5%.The original version of this article supplied to AIP Conference Proceedings contained an error in the writing of the formulas (no. 1-15). The paper originally appeared as, double formulas, but the correct writing is single formulas (no. 1-15). An updated version of this article, with the corrected formulas was published on 29 November 2018.