14 Predictive Modeling of Tox21 Data
281
response induced by certain chemicals or chemical classes are established, which
might be predictive of adverse health outcomes in humans.
Aiming to identify in vitro chemical signatures that could act as predictive surrogates for in vivo toxicity, the US Tox21 (toxicology in the twenty-first century)
program was established in 2008 with an emphasis on developing new methodologies to evaluate the potential risk of thousands of environmental chemicals on
human health [2–5]. The Tox21 program is a collaboration involving the National
Toxicology Program (NTP) at the National Institute of Environmental Health Sciences (NIEHS), the National Center for Computational Toxicology (NCCT) at the
US Environmental Protection Agency (EPA), the intramural research division of the
National Center for Advancing Translational Sciences (NCATS), and the US Food
and Drug Administration (FDA). The goal of the Tox21 program is threefold:
(1) analyze the patterns of compound-induced biological responses in order to identify toxicity pathways and compound mechanisms of toxicity;
(2) prioritize compounds for further extensive toxicological evaluation; and
(3) develop predictive models for biological response in human beings.
This inter-federal agency partnership established a collection of ~10,000 environmental chemicals and drugs (Tox21 10K library) to profile for potential effects
on human health [6, 7]. The Tox21 10K library is screened against a large panel of
cell-based assays in a quantitative high-throughput screening (qHTS) format as 15-pt
titration series in triplicate [8]. qHTS generates a concentration–response curve for
each compound, which greatly reduces the frequency of false positives and negatives.
During phase II (2011–2017), the production phase of the program, the Tox21 10K
library had been screened against a panel of more than 50 assays with an initial focus
on the nuclear receptor (NR) [9–13] and stress response (SR) pathways (Table 14.1)
[14–16] in qHTS format producing over 85 million data points to date [7, 17–19].
The results form a rich set of compound in vitro activity profiles that can serve as
the basis for mechanism of compound toxicity hypotheses generation and predictive
modeling. However, as with any new technology, the reliability and the relevance of
the approach need to be evaluated and validated. Here, we describe a few examples
using this dataset to build in silico models to predict in vivo including human toxicity.
The performances of in vitro assay data-based models are compared with those of
chemical structure information, animal toxicity data, and literature annotations on
compound target and mode of action. Strategies are proposed to optimize the biological space coverage of in vitro assays in order to establish comprehensive compound
activity profiles for improved toxicity prediction.
281
response induced by certain chemicals or chemical classes are established, which
might be predictive of adverse health outcomes in humans.
Aiming to identify in vitro chemical signatures that could act as predictive surrogates for in vivo toxicity, the US Tox21 (toxicology in the twenty-first century)
program was established in 2008 with an emphasis on developing new methodologies to evaluate the potential risk of thousands of environmental chemicals on
human health [2–5]. The Tox21 program is a collaboration involving the National
Toxicology Program (NTP) at the National Institute of Environmental Health Sciences (NIEHS), the National Center for Computational Toxicology (NCCT) at the
US Environmental Protection Agency (EPA), the intramural research division of the
National Center for Advancing Translational Sciences (NCATS), and the US Food
and Drug Administration (FDA). The goal of the Tox21 program is threefold:
(1) analyze the patterns of compound-induced biological responses in order to identify toxicity pathways and compound mechanisms of toxicity;
(2) prioritize compounds for further extensive toxicological evaluation; and
(3) develop predictive models for biological response in human beings.
This inter-federal agency partnership established a collection of ~10,000 environmental chemicals and drugs (Tox21 10K library) to profile for potential effects
on human health [6, 7]. The Tox21 10K library is screened against a large panel of
cell-based assays in a quantitative high-throughput screening (qHTS) format as 15-pt
titration series in triplicate [8]. qHTS generates a concentration–response curve for
each compound, which greatly reduces the frequency of false positives and negatives.
During phase II (2011–2017), the production phase of the program, the Tox21 10K
library had been screened against a panel of more than 50 assays with an initial focus
on the nuclear receptor (NR) [9–13] and stress response (SR) pathways (Table 14.1)
[14–16] in qHTS format producing over 85 million data points to date [7, 17–19].
The results form a rich set of compound in vitro activity profiles that can serve as
the basis for mechanism of compound toxicity hypotheses generation and predictive
modeling. However, as with any new technology, the reliability and the relevance of
the approach need to be evaluated and validated. Here, we describe a few examples
using this dataset to build in silico models to predict in vivo including human toxicity.
The performances of in vitro assay data-based models are compared with those of
chemical structure information, animal toxicity data, and literature annotations on
compound target and mode of action. Strategies are proposed to optimize the biological space coverage of in vitro assays in order to establish comprehensive compound
activity profiles for improved toxicity prediction.