Machine learning model for antiproliferative virtual screening of herbal compounds against hepatocellular carcinoma

Abstract

Machine learning (ML) has been applied to virtual screening in discovering novel antiproliferative agents against hepatocellular carcinoma (HCC) from bioactive herbal compounds. ML models that have been performed to predict activities were constructed using extended connectivity fingerprints up to four bonds (ECFP4) to represent molecule structures. The dataset, consisting of 5,460 molecules with antiproliferative activity against HepG2, was obtained from the ChEMBL database. An evaluation of several algorithms on the primary dataset revealed that random forest gave the best model, both regression and classification performance, with coefficient determinations of 0.803 and 0.954, respectively. Virtual screening results identified some bioactive compounds from the medicinal plants that are predicted to have potential activities as antiproliferation. This action can reduce the number of chemical samples that will be tested in the wet lab using HepG2 cells as an in vitro assay model. Along these lines, this action will reduce trial and error so there will be more function in time, cost, and exertion in disclosing anticancer medications.