Tox21Challenge to Build Predictive Models of Nuclear Receptor and Stress Response Pathways as Mediated by Exposure to Environmental Chemicals and Drugs

Abstract

Tens of thousands of chemicals with poorly understood biological properties are released into the environment each day. High-throughput screening (HTS) is potentially a more efficient and cost-effective alternative to traditional toxicity tests. Using HTS, one can profile chemicals for potential adverse effects and prioritize a manageable number for more in-depth testing. Importantly, it can provide clues to mechanism of toxicity. The Tox21 program has generated >50 million quantitative high-throughput screening (qHTS) data points. A library of several thousands of compounds, including environmental chemicals and drugs, is screened against a panel of nuclear receptor (NR) and stress response (SR) pathway assays. The National Center for Advancing Translational Sciences (NCATS) has organized an international data challenge in order to "crowd-source" data and build predictive toxicity models. This Challenge asks a "crowd" of researchers to use these data to elucidate the extent to which the interference of biochemical and cellular pathways by compounds can be inferred from chemical structure data. The data generated against the Tox21 library served as the training set for this modeling Challenge. The competition attracted participants from 18 different countries to develop computational models aimed at better predicting chemical toxicity. The winning models from nearly 400 model submissions all achieved >80% accuracy. Several models exceeded 90% accuracy, which was measured by area under the receiver operating characteristic curve (AUC-ROC). Combining the winning models with the knowledge already gained from Tox21 screening data are expected to improve the community's ability to prioritize novel chemicals with respect to potential human health concern.

Predicting drug properties with parameter-free machine learning: pareto-optimal embedded modeling (POEM)

“…The Androgen Receptor (AR) activity dataset was taken from the Tox21 Challenge, a federal collaboration involving NIH, EPA and the FDA aimed to develop better toxicity assessment methods. For this sub-challenge, 31 teams contributed different predictive models, with the leading ROC AUC at 0.828 ( 37 ). Consistently, POEM matched or outperformed expert-developed models reported in the literature.…”

Deep Learning Based Regression and Multiclass Models for Acute Oral Toxicity Prediction with Automatic Chemical Feature Extraction

“…Recently, Duvenaud et al 29 and Kearnes et al 30 first used CNNs to successfully implement similar source-based methods. The stateof-the-art performance on some public datasets [31][32][33][34][35][36] suggests that molecular graph encoding (MGE) methods based on multiple NNs have potential in the field of chemoinformatics. In principal, MGE is an ideal representation of chemical structures without information loss.…”

ToxicBlend: Virtual screening of toxic compounds with ensemble predictors

“…As witnessed in the Tox21 Challenge, a spin-off of the ToxCast project organized in 2014 [6], techniques based around deep neural networks (DNNs), are able to provide state-of-the-art performance for virtual toxicity screening [7][8][9][10][11][12]. The winning solution of [13] utilized an ensemble of various classifiers built upon many thousands of chemical descriptors.…”