Hazard assessment is the first step in GreenScreen® for Safer Chemicals. Among the 18 human health and environmental hazard endpoints, carcinogenicity is one of the most hazardous which can make a compound to be of high concern for replacement. To help predict the carcinogenicity for new compounds in development, we develop QSAR (quantitative structure-activity relationship) models based on 25338<br/>compounds classified H (High), M (Medium) and L (low) in GreenScreen Hazard Assessment for carcinogenicity. The descriptors contain either all the descriptors calculated by PaDEL or the 881-bit PubChem substructure fingerprints. We use SVM (support vector machine), C5.0 decision tree, RF (random forest) and GCN (Graph Convolutional Network) to build the models. 10-fold cross validation is performed<br/>and we also perform external validation. The applicability domain is defined using confidence or k-NN methods to evaluate the distance between training set and the compound to be predicted. Data from Carcinogenic Potency Database (CPDB) will be<br/>used in external validation. The construction and validation of our model obeys the OECD principles for the Validation, for Regulatory Purpose, of (Q)SAR Models. We hope our model can help to assess the carcinogenicity of new compounds for the<br/>green chemistry and sustainability.

QSAR model of carcinogenicity based on data from GreenScreen hazard assessment

Fall 2022 abstracts, posters, and presentations will launch in mid-September.

ACS Fall 2022

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.