374
E. Benfenati et al.
Forty-eight different models run together in order to cover all these endpoints.
As in the case of VEGA, it happens that for one endpoint more models exist. In
this case, the multiple results are integrated into a unique value. The software generates microbial degradation products, which may occur in the environment, based on
the program developed by the University of Minneapolis, and implemented within
EnviPath (https://envipath.org/). It contains 235 separate degradation processes, identifying degradation products. JANUS evaluates the PBT, CMR, and ED properties
of the parental compound and of the degradation products automatically.
The PBT and CMR properties refer to the European regulation, and thus, a chemical is considered PBT only if it is persistent, bioaccumulative, and toxic at the same
time, according to the threshold values defined by REACH [11] and the regulation
for the Classification, Labeling, and Packaging (CLP) [18]. Thus, for persistence,
the software looks for persistence in water, sediment, and soil. A chemical is defined
persistent if it is persistent in at least one of these three compartments. The evaluation
for the ecotoxicity refers to aquatic toxicity only. It covers fish, daphnia, and algae,
including both acute and chronic toxicity. A chemical is toxic if it is toxic for at least
one of the three trophic levels. For fish acute toxicity, there are both general and
specie-specific models. For bioaccumulation, the bioconcentration factor in fish is
modeled.
In case of mutagenicity, the system covers the Ames test. For carcinogenicity, the
system uses both regression and classifier models, in order to provide a quantitative
value, related to the slope factor, both for oral and inhalation exposure. Also for
reprotoxicity, there are regression and classifier models.
ED is covered considering general models for ED and specific models for estrogen
and androgen receptors.
For each property addressed within the PBT, CMR, and ED, there are workflows
integrating the results of multiple models, integrating both experimental (retrieved by
the models) and predicted values. There are further models supporting the assessment.
For instance, the workflow for BCF also verifies if the logP value is consistent with the
BCF value. In case of aquatic toxicity, the water solubility is checked and compared
with the toxicity value. For mutagenicity, the metabolism according to the S9 fraction
is also considered.
The user can provide to the system experimental values, if available. Then, for
P and B only, the workflow first refers to a series of families identified as priority
pollutants. If the target compound belongs to one of these families, the score related
to the family is assigned.
The components to assign the final score for the prioritization list refer to subscore related to the individual properties and endpoints assessed. As a general rule,
experimental values have higher reliability than predicted values; multiple concordant values have higher reliability than single values; and the spread of the values in
case of multiple values decreases the reliability.
E. Benfenati et al.
Forty-eight different models run together in order to cover all these endpoints.
As in the case of VEGA, it happens that for one endpoint more models exist. In
this case, the multiple results are integrated into a unique value. The software generates microbial degradation products, which may occur in the environment, based on
the program developed by the University of Minneapolis, and implemented within
EnviPath (https://envipath.org/). It contains 235 separate degradation processes, identifying degradation products. JANUS evaluates the PBT, CMR, and ED properties
of the parental compound and of the degradation products automatically.
The PBT and CMR properties refer to the European regulation, and thus, a chemical is considered PBT only if it is persistent, bioaccumulative, and toxic at the same
time, according to the threshold values defined by REACH [11] and the regulation
for the Classification, Labeling, and Packaging (CLP) [18]. Thus, for persistence,
the software looks for persistence in water, sediment, and soil. A chemical is defined
persistent if it is persistent in at least one of these three compartments. The evaluation
for the ecotoxicity refers to aquatic toxicity only. It covers fish, daphnia, and algae,
including both acute and chronic toxicity. A chemical is toxic if it is toxic for at least
one of the three trophic levels. For fish acute toxicity, there are both general and
specie-specific models. For bioaccumulation, the bioconcentration factor in fish is
modeled.
In case of mutagenicity, the system covers the Ames test. For carcinogenicity, the
system uses both regression and classifier models, in order to provide a quantitative
value, related to the slope factor, both for oral and inhalation exposure. Also for
reprotoxicity, there are regression and classifier models.
ED is covered considering general models for ED and specific models for estrogen
and androgen receptors.
For each property addressed within the PBT, CMR, and ED, there are workflows
integrating the results of multiple models, integrating both experimental (retrieved by
the models) and predicted values. There are further models supporting the assessment.
For instance, the workflow for BCF also verifies if the logP value is consistent with the
BCF value. In case of aquatic toxicity, the water solubility is checked and compared
with the toxicity value. For mutagenicity, the metabolism according to the S9 fraction
is also considered.
The user can provide to the system experimental values, if available. Then, for
P and B only, the workflow first refers to a series of families identified as priority
pollutants. If the target compound belongs to one of these families, the score related
to the family is assigned.
The components to assign the final score for the prioritization list refer to subscore related to the individual properties and endpoints assessed. As a general rule,
experimental values have higher reliability than predicted values; multiple concordant values have higher reliability than single values; and the spread of the values in
case of multiple values decreases the reliability.