S3: ANN-based QNTR model development

ANN-based QNTR model development using PChem score-based data set

S4: ANN-based QNTR model development, OpenTox Asia 2017

Hyung Gi Byun


Kangwon National University




Modeling QNTR (quantitative nanoproperty-toxicity relationship) and interpreting its result using the integrated database, which contains a lot of toxic experiment results obtained from several sources such as article and toxic experiments in vivo and in vitro, are one of the current challenges.In terms of data quality management of the integrated database, missing value, type, and noise decrease the level of quality of the database. In another aspect regarding QNTR model development, various descriptors and different toxic experimental conditions within the integrated database make it difficult to implement and interpret the model.

For coping with the problem regarding data quality of the database, PChem score-based data screening method has been applied to the database by Prof. Yoon’s team at Hanyang University in Korea. Using their database, we have been developed ANN-based QNTR prediction models which are capable of using different variable types such as numeric attribute and nominal attribute. More specifically, six Rprop-based QNTR models were developed using six data sets selected by the data screening method. In this talk, we present how we have been developed ANN-based models using PChem score-based data sets and how we can interpret the results of the developed QNTR models.