You are here:
KC-QSARs: Modeling Key Characteristics of Chemical Carcinogenesis
Citation:
Borrel, A., R. Rudel, K. Houck, AND N. Kleinstreuer. KC-QSARs: Modeling Key Characteristics of Chemical Carcinogenesis. International Conference on Environmental Mutagens, Ottawa, N/A, CANADA, August 28 - September 02, 2021.
Impact/Purpose:
Abstract submitted for a presentation to the 13th International Conference on Environmental Mutagens: Maintaining Genomic Health in a Changing World September 2021. A crucial challenge in the environmental health field is rapidly screening chemicals for their potential impact on in vivo human carcinogenesis using in silico approaches incorporating in vitro or/and structural information. Mechanistic analyses have revealed 10 key characteristics (KCs), such as ‘is genotoxic,’ ‘is immunosuppressive’ or ‘modulates receptor-mediated effects’, that are commonly exhibited by known or suspected carcinogens. Here we describe the development of Quantitative Structure-Activity Relationship (QSAR) models for the KCs of carcinogens (KC-QSARs) from data extracted from the U.S. EPA’s invitroDBv3 databaseFinally, these QSAR models were applied to the DSSTox Database of >800,000 chemicals, and predictions for chemicals with and without in vivo cancer data are discussed. This work provides in silico tools to correlate the in vitro bioactivity response patterns of certain chemicals to their in vivo carcinogenic potential and supports the use of mechanistic data in carcinogen hazard prediction.
Description:
A crucial challenge in the environmental health field is rapidly screening chemicals for their potential impact on in vivo human carcinogenesis using in silico approaches incorporating in vitro or/and structural information. Mechanistic analyses have revealed 10 key characteristics (KCs), such as ‘is genotoxic,’ ‘is immunosuppressive’ or ‘modulates receptor-mediated effects’, that are commonly exhibited by known or suspected carcinogens. Here we describe the development of Quantitative Structure-Activity Relationship (QSAR) models for the KCs of carcinogens (KC-QSARs) from data extracted from the U.S. EPA’s invitroDBv3 database. In the first step, we updated the mapping employed by the International Agency for Research on Carcinogens (IARC) of the Tox21/ToxCast assays to the KCs, using the gene and pathway targets and an expert-driven approach. The Tox21/ToxCast programs screen thousands of chemicals across hundreds of assays in a high-throughput format, providing broad coverage of chemical and biological space. By combining activity results from the mapped assays for each KC, we established response scores (SKC1, SKC2, …) for all tested chemicals, a subset of which had corresponding in vivo cancer bioassay data that was used to establish confidence in the mapping and scoring systems. QSAR models were developed using various machine learning algorithms (Support Vector Machine, Neural Network, Random Forest…).to learn and predict these scores using sets of 1D and 2D molecular descriptors. Finally, these QSAR models were applied to the DSSTox Database of >800,000 chemicals, and predictions for chemicals with and without in vivo cancer data are discussed. This work provides in silico tools to correlate the in vitro bioactivity response patterns of certain chemicals to their in vivo carcinogenic potential and supports the use of mechanistic data in carcinogen hazard prediction. The views expressed in this abstract are those of the authors and do not necessarily represent the views or policies of the U.S. EPA.