38
R. Williams et al.
EC3
Effective concentration to cause a threefold increase in T-cell proliferation
GHS
Globally harmonised system
GPMT Guinea pig maximisation test
h-CLAT Human cell line activation test
IATA
Integrated approach to testing and assessment
KE
Key event
kNN
k-Nearest neighbours
LLNA Local lymph node assay
MIE
Molecular initiating event
MW
Molecular weight
OATP
Organic anion transporting polypeptide
OECD Organisation for Economic Co-operation and Development
PPAR
Peroxisome proliferator-activated receptor
3.1 Introduction
Structural alerts and quantitative structure–activity relationships (QSARs) have a
long history of utility for the qualitative prediction of toxicity. One of the earliest
examples is the Ashby–Tennant superstructure, a chemical concatenation of toxicophores associated with (and causal for) mutagenic and carcinogenic activity [1].
Where such toxicophores were found within a chemical, then that chemical could be
predicted to be mutagenic and carcinogenic. The success of that early in papyro/in
cerebro model led to the development of increasingly sophisticated computational
systems for predicting toxicity in silico which have sought to replicate human-like
reasoning [2, 3] or utilise advanced machine-learning techniques [4, 5].
3.2 Lessons Learnt from Successful Models
Retrospectively, it has become clear that predictions of toxicity from in silico models
have often shown the highest levels of acceptance and application where the endpoints
being modelled are governed by one (or few) molecular initiating events (MIE)
and there are limited absorption, distribution, metabolism, and excretion (ADME)
considerations. An MIE describes the interaction between a chemical and a biological
target and is linked to a toxicity endpoint via an adverse outcome pathway (AOP)
[6]. For endpoints such as skin sensitisation and mutagenicity, toxicity is largely
driven by chemical reactivity, leading to the formation of protein or DNA adducts
which are the respective MIEs. Although metabolic activation and deactivation can
be a precursor to these MIEs, there are relatively few ADME factors to consider.
Thus, in these cases, a model where the activity of a wide range of chemicals can
be assessed (i.e. a global model) can be generated from a single descriptor set (often
chemical fragments) which provides an acceptable simulation of biological reality.
R. Williams et al.
EC3
Effective concentration to cause a threefold increase in T-cell proliferation
GHS
Globally harmonised system
GPMT Guinea pig maximisation test
h-CLAT Human cell line activation test
IATA
Integrated approach to testing and assessment
KE
Key event
kNN
k-Nearest neighbours
LLNA Local lymph node assay
MIE
Molecular initiating event
MW
Molecular weight
OATP
Organic anion transporting polypeptide
OECD Organisation for Economic Co-operation and Development
PPAR
Peroxisome proliferator-activated receptor
3.1 Introduction
Structural alerts and quantitative structure–activity relationships (QSARs) have a
long history of utility for the qualitative prediction of toxicity. One of the earliest
examples is the Ashby–Tennant superstructure, a chemical concatenation of toxicophores associated with (and causal for) mutagenic and carcinogenic activity [1].
Where such toxicophores were found within a chemical, then that chemical could be
predicted to be mutagenic and carcinogenic. The success of that early in papyro/in
cerebro model led to the development of increasingly sophisticated computational
systems for predicting toxicity in silico which have sought to replicate human-like
reasoning [2, 3] or utilise advanced machine-learning techniques [4, 5].
3.2 Lessons Learnt from Successful Models
Retrospectively, it has become clear that predictions of toxicity from in silico models
have often shown the highest levels of acceptance and application where the endpoints
being modelled are governed by one (or few) molecular initiating events (MIE)
and there are limited absorption, distribution, metabolism, and excretion (ADME)
considerations. An MIE describes the interaction between a chemical and a biological
target and is linked to a toxicity endpoint via an adverse outcome pathway (AOP)
[6]. For endpoints such as skin sensitisation and mutagenicity, toxicity is largely
driven by chemical reactivity, leading to the formation of protein or DNA adducts
which are the respective MIEs. Although metabolic activation and deactivation can
be a precursor to these MIEs, there are relatively few ADME factors to consider.
Thus, in these cases, a model where the activity of a wide range of chemicals can
be assessed (i.e. a global model) can be generated from a single descriptor set (often
chemical fragments) which provides an acceptable simulation of biological reality.