In-silico drug screening and potential target identification for hepatocellular carcinoma using Support Vector Machines based on drug screening result

Abstract

Hepatocellular carcinoma (HCC) is a severe liver malignancy with few drug treatment options. In finding an effective treatment for HCC, screening drugs that are already FDA-approved will fast track the clinical trial and drug approval process. Connectivity Map (CMap), a large repository of chemical-induced gene expression profiles, provides the opportunity to analyze drug properties on the basis of gene expression. Support Vector Machines (SVM) were utilized to classify the effectiveness of drugs against HCC using gene expression profiles in CMap. The results of this classification will help us (1) identify genes that are chemically sensitive, and (2) predict the effectiveness of remaining chemicals in CMap in the treatment of HCC and provide a prioritized list of possible HCC drugs for biological verification. Four HCC cell lines were treated with 146 distinct chemicals, and cell viability was examined. SVM successfully classified the effectiveness of the chemicals with an average Area Under ROC Curve (AUROC) of 0.9. Using reported HCC patient samples, we identified chemically sensitive genes that may be possible HCC therapeutic targets, including MT1E, MYC, and GADD45B. Using SVM, several known HCC inhibitors, such as geldanamycin, alvespimycin (HSP90 inhibitors), and doxorubicin (chemotherapy drug), were predicted. Seven out of the 23 predicted drugs were cardiac glycosides, suggesting a link between this drug category and HCC inhibition. The study demonstrates a strategy of in silico drug screening with SVM using a large repository of microarrays based on initial in vitro drug screening. Verifying these results biologically would help develop a more accurate chemical sensitivity model.