Development of machine learning tool for drug class prediction

Abstract

newline vi newlineABSTRACT newlineAdvancement in technology has revolutionized the field of computer science with its vital role in capturing, analyzing, and managing data for better monitoring, understanding, and decision making. The use of information technology in different fields has transformed the traditional approach, and now most of the work is aided with computers in several fields/domains. Computational approaches are also found to be successful in different challenging problems, mainly when a large volume of data is associated with them. The popularity of computational technologies such as artificial intelligence, machine learning, data science, and data mining has also increased phenomenally due to immense applicability, development, and availability of open source technology, and data availability. Although several fields are possessing significant challenges to computational technologies because of their complex nature, such as biology and medicine. The first such problem considered in current work is failure/withdrawn of the drug after market launch. Failure of drug after-market launch is a complex issue, and a high amount of safety, regulatory, chemical and biological data is also associated with it. In previous years, most cases in the post-market failure of drugs have raised serious safety issues. In history, failed drugs had made an epidemic that was responsible for thousands of deaths together with economic loss. newlineThe second problem which was considered in current work is the prediction of the multi-disease potential of drug-like molecules. The drug which can treat multiple diseases is always like a dream come true. Most of the drugs which are currently approved and present in the market are used to treat one particular disease, but few drugs are recommended to treat different diseases (more than one disease). Such drugs can be the better candidate for the treatment of complex diseases like cancer obesity, Alzheimer etc. newlineDevelopment of computational methods to predict future post-market failure of molecules that are currently in drug discovery pipeline can save human health/life and economic losses associated with drug failure. A computational machine learning-based method was developed from the information of all approved and failed drugs with reasonably high accuracy on independent test data to justify its reliability and usefulness. The developed method is implemented on a web server at http://117.242.138.233/chem/. newlinevii newlineSimilarly, the first-ever machine learning-based computational method was successfully developed for the prediction of the multi-disease potential of a drug-like molecule from their 2D or 3D structure. The method can be used before any laboratory experiments to check the multi-disease potential of any drug-like molecule. Relatively high accuracy of methods in training and independent test set justified its reliability and usefulness. The developed method is implemented on the webserver for its unrestricted usage in the research community (available at http://117.242.138.233/chem1/), which is expected to speed up the discovery of multi-disease drugs. The method is also expedited the drug repurposing process. Keyword: Withdrawn drug, machine learning, chemoinformatics, drug design, bioinformatics, Artificial intelligence,