POSTER ABSTRACT / DETAILS:
Recently, in silico evaluation of drug toxicity has been issued because of its economic feasibility and speed. Up to now, many different types of algorithms have been applied to construct precise evaluation of toxicity by analyzing physic-chemical properties of drugs. However, there is no benchmarking analysis of applied algorithms by comparing performance using same data set. We had performed benchmark analysis of widely used 4 classification algorithms, KNN, LDA, SVM, and ANN. For benchmark analysis, we have extracted molecular signatures of 155 frequently used drugs using Mold2.
For training, the toxicity information of 155 drugs is also retrieved from The Comparative Toxicogenomics Database (CTD) and each drug is sub-grouped by their organ toxicity. The performance of applied algorithms was evaluated using LOOCV. From the benchmark analysis, we have found LDA and SVM are better than other algorithms for structure-based toxicity evaluation.