Modern Chemistry & Applications

Prediction of serious adverse events using machine learning

Joint Event on 19^th International Conference on Medicinal Chemistry & Multi Targeted Drug Delivery & International Conference on Catalysis and Pyrolysis

November 05-06, 2018 | San Francisco, USA

Tatsuya Takagi

Osaka University, Japan

Keynote: Mod Chem Appl

Abstract:

As is well known, adverse events of drugs are inevitable. However, in the case of serious (sometimes fatal) adverse events, certainly, they exert some negative impacts on quality of life for the patients. Although fortunately, such adverse events rarely occur, there are some drugs which cause such events relatively frequently. We consider that there must be some features of chemical structures as well as quantum chemical indices for the drugs. We utilized several machine learning methods including logistic regressions, support vector machines and random forests. As the serious adverse events, rhabdomyolysis, Stevens- Johnson syndrome, toxic epidermal necrolysis, malignant syndrome, posterior reversible encephalopathy, leukoencephalopathy and long QT syndrome were dealt with. Here, firstly, we tried to predict the drugs with the higher possibility of the malignant syndrome. Data were extracted from JADER (Japanese Adverse Drug Event Report database) constructed by PMDA (Pharmaceutical and Medical Device Agency). The predictabilities were expressed using AUC (Area Under the Curve) of ROC curves and percentages of correct answers (ACC). As explanation variables, chemical descriptors generated by ???Mordred???, which was developed by our group, ATCC (Anatomical Therapeutic Chemical Classification system) and some physicochemical ones computed by MOE (by CCG) were adopted. Although the 1st model showed 0.80 (AUC) and 0.70 (ACC), 0.91 and 0.86 (AUC and ACC, respectively) were obtained in the case of the best model using MV (Majority Voting), after selecting the significant descriptors. As the descriptors for the model, MATS8se (moran coefficient of lag 8 weighted by Sanderson EN) and Nitrogen atom having three single bonds were especially significant. Drug designers should give attention to the two descriptors.

Biography :

Tatsuya Takagi has completed his PhD at the age of 32 from Osaka University. At that time, he had been an Assistant Professor of School of Pharmaceutical Sciences, Osaka University for 5 years. Then, since 1993, he had worked for the Genome Information Research Center, Osaka University as an Associate Professor until he became a Professor of Graduate School of Pharmaceutical Sciences, Osaka University in 1998. He has published more than 150 papers in reputed journals and had served as Chairman of Division of Structure-Activity Relationship of the Pharmaceutical Society of Japan for three years (until March 2017).

E-mail: ttakagi@phs.osaka-u.ac.jp