366
E. Benfenati et al.
Abbreviations
AD
Applicability domain
ADI Applicability domain index
BCF Bioconcentration factor
CLP Classification, labeling, and packaging
CMR Carcinogenic, mutagenic, or reprotoxicants
ED
Endocrine disruptors
kNN k-nearest neighbor
NTM Non-testing methods
PBT Persistent, bioaccumulative, and toxic
SA
Structural alerts
SOM Self-organizing map
WoE Weight-of-evidence
18.1 Introduction
In the last decades, dramatic changes occurred in the field of non-testing methods
(NTM). NTM include QSAR models and read-across. These two areas moved from
quite different perspectives and uses. Indeed, QSAR originally was a model developed for research purposes, to evaluate the reasons for the effects within specific
chemical categories. The possible relationship between the chemical structure and
the property value, in particular the toxicological and ecotoxicological properties,
has been studied for decades. In the case of the ecotoxicological properties, most
of the studies address certain chemical families and most commonly aquatic acute
toxicity [1]. Generally, for these properties, regression models have been developed.
In the case of toxicological properties, a number of studies addressed categorical
methods to discriminate toxic versus non-toxic compounds. Indeed, many studies
focused on mutagenicity via the Ames test [2].
For both toxicological and ecotoxicological studies, initially the assumption was
that the model refers to a certain chemical family. In the case of ecotoxicity, the model
was applicable to the family, and in the same way, for toxicological categorical
models, the effect was associated with the fact that the chemical belongs to the
family. Thus, initially, the attention started from the identification of a certain family
of compounds, and the reasoning affected this local situation.
In a second phase, the studies explored the possibilities to extend the chemical
domain of the model. One approach was simply to develop collections of sub-models.
However, the availability of software to calculate a very large number of descriptors, of more sophisticated algorithms, and of more powerful computers at low cost
resulted in a dramatic increase of models able to cover many families of compounds
for many (eco)toxicological properties. At this point, the QSAR model was in principle a general model [3].
E. Benfenati et al.
Abbreviations
AD
Applicability domain
ADI Applicability domain index
BCF Bioconcentration factor
CLP Classification, labeling, and packaging
CMR Carcinogenic, mutagenic, or reprotoxicants
ED
Endocrine disruptors
kNN k-nearest neighbor
NTM Non-testing methods
PBT Persistent, bioaccumulative, and toxic
SA
Structural alerts
SOM Self-organizing map
WoE Weight-of-evidence
18.1 Introduction
In the last decades, dramatic changes occurred in the field of non-testing methods
(NTM). NTM include QSAR models and read-across. These two areas moved from
quite different perspectives and uses. Indeed, QSAR originally was a model developed for research purposes, to evaluate the reasons for the effects within specific
chemical categories. The possible relationship between the chemical structure and
the property value, in particular the toxicological and ecotoxicological properties,
has been studied for decades. In the case of the ecotoxicological properties, most
of the studies address certain chemical families and most commonly aquatic acute
toxicity [1]. Generally, for these properties, regression models have been developed.
In the case of toxicological properties, a number of studies addressed categorical
methods to discriminate toxic versus non-toxic compounds. Indeed, many studies
focused on mutagenicity via the Ames test [2].
For both toxicological and ecotoxicological studies, initially the assumption was
that the model refers to a certain chemical family. In the case of ecotoxicity, the model
was applicable to the family, and in the same way, for toxicological categorical
models, the effect was associated with the fact that the chemical belongs to the
family. Thus, initially, the attention started from the identification of a certain family
of compounds, and the reasoning affected this local situation.
In a second phase, the studies explored the possibilities to extend the chemical
domain of the model. One approach was simply to develop collections of sub-models.
However, the availability of software to calculate a very large number of descriptors, of more sophisticated algorithms, and of more powerful computers at low cost
resulted in a dramatic increase of models able to cover many families of compounds
for many (eco)toxicological properties. At this point, the QSAR model was in principle a general model [3].