376
E. Benfenati et al.
case, VEGA, within the ADI program, indicates if there are fragments in common
with the target compound, and/or fragments present only in the source compound.
This is a very useful feature because it helps with reasoning about the real similarity
between the source and target compounds, beyond chemical consideration. Indeed,
the SA is endpoint specific, and if the target and the source compounds share the
same SA, this increases the similarity, but if the source compound has a different SA,
it can be responsible for the difference in the property values between the source and
the target compound.
18.3.2 Other Programs for Read-Across
Above we mentioned all models available within VEGA can be used for read-across,
simply by looking at similar compounds, and possibly at their SA. In addition, some
models within VEGA are based on the k-nearest neighbor (kNN) algorithm. This
approach is a kind of automatic read-across, which makes predictions based on
similar compounds. Also in this case, we recommend looking at the results of the
other models, in particular if there are models indicating SA. This because it may
be that all the most similar compounds to the target chemical contain a SA which is
responsible for the effect, but it is not present in the source compound.
The idea to carefully consider SA is at the basis of ToxRead. In this case, the
software does the evaluation automatically. This increases the reproducibility of
the results offered by the read-across tool. It has been shown that ToxRead is more
reproducible and less affected by the subjectivity of the read-across evaluation, which
is a major drawback of read-across [26]. This because ToxRead indicates a wide series
of SA and not only those initially identified by the user.
ToxDelta may be used joined with ToxRead. Its basis is different from ToxRead,
since ToxRead refers to the similarity (chemical and toxicological/environmental,
through the SA and the other features it covers), while ToxDelta focuses on the
dissimilarity. ToxRead looks for several similar compounds, while ToxDelta should
be used for the substances to be compared.
18.3.3 The JANUS Software
JANUS further proceeds along the software PROMETHEUS for prioritization. It
includes a high number of models, which will be also made available as individual tools within VEGA. The novelty of JANUS is that the prioritization covers a
very large set of properties, related to a series of endpoints identified by regulation.
Previous tools for prioritization, including PROMETHEUS, addressed mainly PBT.
PROMETHEUS evaluated the T only through fish toxicity and did not give a numeric
evaluation of each property but only a classification. Other software, like the PBT
index [27], gives a PBT evaluation but not the three properties separately. Again,
E. Benfenati et al.
case, VEGA, within the ADI program, indicates if there are fragments in common
with the target compound, and/or fragments present only in the source compound.
This is a very useful feature because it helps with reasoning about the real similarity
between the source and target compounds, beyond chemical consideration. Indeed,
the SA is endpoint specific, and if the target and the source compounds share the
same SA, this increases the similarity, but if the source compound has a different SA,
it can be responsible for the difference in the property values between the source and
the target compound.
18.3.2 Other Programs for Read-Across
Above we mentioned all models available within VEGA can be used for read-across,
simply by looking at similar compounds, and possibly at their SA. In addition, some
models within VEGA are based on the k-nearest neighbor (kNN) algorithm. This
approach is a kind of automatic read-across, which makes predictions based on
similar compounds. Also in this case, we recommend looking at the results of the
other models, in particular if there are models indicating SA. This because it may
be that all the most similar compounds to the target chemical contain a SA which is
responsible for the effect, but it is not present in the source compound.
The idea to carefully consider SA is at the basis of ToxRead. In this case, the
software does the evaluation automatically. This increases the reproducibility of
the results offered by the read-across tool. It has been shown that ToxRead is more
reproducible and less affected by the subjectivity of the read-across evaluation, which
is a major drawback of read-across [26]. This because ToxRead indicates a wide series
of SA and not only those initially identified by the user.
ToxDelta may be used joined with ToxRead. Its basis is different from ToxRead,
since ToxRead refers to the similarity (chemical and toxicological/environmental,
through the SA and the other features it covers), while ToxDelta focuses on the
dissimilarity. ToxRead looks for several similar compounds, while ToxDelta should
be used for the substances to be compared.
18.3.3 The JANUS Software
JANUS further proceeds along the software PROMETHEUS for prioritization. It
includes a high number of models, which will be also made available as individual tools within VEGA. The novelty of JANUS is that the prioritization covers a
very large set of properties, related to a series of endpoints identified by regulation.
Previous tools for prioritization, including PROMETHEUS, addressed mainly PBT.
PROMETHEUS evaluated the T only through fish toxicity and did not give a numeric
evaluation of each property but only a classification. Other software, like the PBT
index [27], gives a PBT evaluation but not the three properties separately. Again,