Artificial neural network modeling of antimycobacterial chemical space to introduce efficient descriptors employed for drug design

Abstract Tuberculosis has become a serious condition with an estimated 2 million deaths each year in the world. According to WHO report, multi-resistant tuberculosis is responsible for approximately 460 thousand recent cases per year and for about 740 thousand patients infected by both Mycobacterium tuberculosis and HIV/AIDS. In the current study, several bioactive structure databases were analyzed using cheminformatics tools to correlate the chemical structures of different compounds with their pharmacological activities; in addition, these tools were tried to identify molecules that could be candidate for experimental assays. In this regard, for defining the effective chemical compounds against Mycobacterium , a database consisting of 400 antimycobacterial compounds has been constructed. In the next step, more than 1400 molecular descriptors were defined by DRAGON application server for each compound. Then, the resulting descriptors were clustered by kNN and k-means clustering methods to be employed for ANN modeling. Utilizing PLS and ANN modeling methods led to building a model for predicting minimum inhibitory concentration (MIC) with R 2  = 0.98 and MSE = 0.0002. Applying the mentioned cheminformatics tools, it would be possible to design and introduce new compounds with broad applications in antimycobacterial drug discovery and development.