Abstract OpenTox Asia 2019 

PASS AS A TOOL FOR IN SILICO PREDICTION OF DRUG ADVERSE & TOXIC EFFECTS

Rajesh K. Goel and Vladimir V. Poroikov, Punjabi University, Patiala, Punjab, India; Institute of Biomedical Chemistry, Moscow, Russia

Application of the predictive toxicology approaches may significantly increase the yield in drug discovery and development. The in vitro and animal-based toxicity testing has a predictive value of about 10 percent [1]. Thus, the improvement of accuracy of in silico predictive methods will reduce or even replace the need for animal testing in the future. Moreover, nowadays regulatory agencies encourage the utilization of in silico data in support of in vivo data for potential hazard identification and characterization [2]. Therefore, we decided to study if the application of PASS (Prediction of Activity Spectra for Substances) [3, 4] may allow identifying the potentially dangerous pharmaceutical agents. PASS applicability to the analysis of putative targets and mechanisms of toxicity was discussed earlier [5]. Current PASS version 2017 predicts 494 adverse & toxic effects with the average accuracy of about 87% based on the structural formulae of drug-like substances. We prepared the manually curated test set included the structures of 276 drug-like molecules withdrawn from the market due to the substantial adverse or toxic actions (Set I). For the comparative analysis, we extracted several other sets from the Clarivate Analytics Integrity database [6]: 312 molecules presently studied in phase II clinical trials (Set II), 165 molecules launched in 2005-2013 (Set III), and 112 molecules launched in 2014-2019 (Set IV). We predicted adverse and toxic effects and calculated the average number (AvN) of predicted activities for all four sets. As a result, we found that AvN values are 311, 224, 179 and 108 for the sets I, II, III and IV, respectively. Thus, the number of predicted adverse & toxic effects for the withdrawn drugs is about one and a half times more than those for the drugs launched in 2005-2013, and almost three times more than those for the drugs launched in 2014-2019. The number of predicted activities for drugs in phase II clinical trials is in between the numbers obtained for the launched and withdrawn drugs. If one chooses AvN=108 as a cutoff value for discrimination between the safety and hazardous drugs, 34 molecules from the Set I (12%) will not be identified as hazardous entities. Thus, the accuracy of prediction in this case study achieved 88%, which demonstrates the applicability of PASS in predictive toxicology. Further development of the proposed approach and its limitations will be discussed.

1. 1. Shanks N. et al. Philosophy, Ethics, and Humanities in Medicine. 2009, 4: 2.
2. 2. Myatt G.J. et al. Regul Toxicol Pharmacol. 2018, 96: 1-17.
3. 3. Filimonov D.A. et al. Chem. Heterocycl. Compnds. 2014, 50: 444-457.
4. 4. PASS Online URL [www.way2drug.com/passonline]
5. 5. Poroikov V.V. et al. SAR & QSAR Environ. Res. 2007, 18: 101-110.
6. 6. Clarivate Analytics Integrity URL [integrity.clarivate.com]