368
E. Benfenati et al.
18.2 Method and Materials
18.2.1 The VEGAHUB Structure
Within the VEGAHUB platform, there are multiple tools, which are dedicated to
the exploration and analysis of the properties of chemical substances. The main
components of VEGAHAB are:
• VEGA
• ToxRead
• ToxWeight
• ToxDelta
• JANUS.
Furthermore, there are links to general tools, which can be used to develop new
models: SARpy [13], VEGA-based tools, SOM (Self-Organizing Map) tool, and
CORAL [14].
18.2.2 The QSAR Models in VEGA
VEGA can be freely downloaded from the website www.vegahub.eu, after registration. VEGA offers a collection of QSAR models. Indeed, there are tens of QSAR
models available within VEGA, addressing physicochemical, environmental, ecotoxicological, and toxicological properties. Table 18.1 lists these models. We are
continuing adding models, and thus, this list is changing rapidly.
There may be more than one model for the same endpoint. This increases the
robustness of the overall prediction and the confidence of the final assessment.
The different models usually have different origin, including training sets of compounds, chemical descriptors, and algorithms. The specific information on each
model, including the chemicals in the training and test sets, is available from VEGA.
Figure 18.1 shows where to find this information.
The user can make the prediction for one single chemical, or for a large set of
compounds, and the software allow both methods of entry, for one or few chemicals
(using the SMILES and an interactive page), or a collection of chemicals in a table.
VEGA automatically checks for the consistency of the SMILES and reports if
there are errors. VEGA transforms the SMILES into its specific SMILES format,
in order to have reproducible results. The results of each model are not simply the
predicted value, but in addition, the most similar chemicals and the evaluation of the
reliability of the prediction are shown.
The most similar compounds are identified using a specific program for similarity
(the algorithm is described in [15]), which has been optimized to balance different
approaches.
E. Benfenati et al.
18.2 Method and Materials
18.2.1 The VEGAHUB Structure
Within the VEGAHUB platform, there are multiple tools, which are dedicated to
the exploration and analysis of the properties of chemical substances. The main
components of VEGAHAB are:
• VEGA
• ToxRead
• ToxWeight
• ToxDelta
• JANUS.
Furthermore, there are links to general tools, which can be used to develop new
models: SARpy [13], VEGA-based tools, SOM (Self-Organizing Map) tool, and
CORAL [14].
18.2.2 The QSAR Models in VEGA
VEGA can be freely downloaded from the website www.vegahub.eu, after registration. VEGA offers a collection of QSAR models. Indeed, there are tens of QSAR
models available within VEGA, addressing physicochemical, environmental, ecotoxicological, and toxicological properties. Table 18.1 lists these models. We are
continuing adding models, and thus, this list is changing rapidly.
There may be more than one model for the same endpoint. This increases the
robustness of the overall prediction and the confidence of the final assessment.
The different models usually have different origin, including training sets of compounds, chemical descriptors, and algorithms. The specific information on each
model, including the chemicals in the training and test sets, is available from VEGA.
Figure 18.1 shows where to find this information.
The user can make the prediction for one single chemical, or for a large set of
compounds, and the software allow both methods of entry, for one or few chemicals
(using the SMILES and an interactive page), or a collection of chemicals in a table.
VEGA automatically checks for the consistency of the SMILES and reports if
there are errors. VEGA transforms the SMILES into its specific SMILES format,
in order to have reproducible results. The results of each model are not simply the
predicted value, but in addition, the most similar chemicals and the evaluation of the
reliability of the prediction are shown.
The most similar compounds are identified using a specific program for similarity
(the algorithm is described in [15]), which has been optimized to balance different
approaches.