2.13.5. pH-Dependent Attributes for Use in Models
Starting from ADMET Predictor X.4, ADMET Predictor now allows for the specification of a pH value for descriptors or models that depend on pH (e.g., fractions ionized, S+S_pH) which will then be used when that descriptor or model is deployed as part of another model. The mechanism for doing so is to name the descriptor/model as <Original name>@<pH value>. For example, if one wanted the log D model (S+logD) to be calculated at a pH value other than 7.4, for example 7.0, in a Risk model, then one should use the name S+logD@7.0 when referring to the model when defining the Risk rules. If one wanted to use S+S_pH at pH 6.5 for use in building a model in ADMET Modeler, the protocol would be as follows. Calculate ADMET properties at pH 6.5 by entering 6.5 in the pH box (1) in the Calculate ADMET properties dialog.
Calculate ADMET properties for the training and test sets used to build the model. Before launching ADMET Modeler, rename the S+S_pH column (using Modify Attribute….) in the spreadsheet as S+S_pH@6.5. Select that descriptor as one of the descriptors for training the model of interest as shown below.
Similarly, if one wishes to use one of the fractions ionized as a descriptor at a particular pH, then one should check on the box “Also calculate ionization descriptors at this pH” and rename the appropriate fraction ionized descriptor(s) by appending @<pH value> to the name.
Note than when this box is checked on it will invalidate the results of the pH dependent models such as S+logD and S+S_pH. If one wishes to use both pH dependent models and fractions ionized at a particular pH in building a model, this needs to be done as a two-step process. First, Calculate ADMET should be performed with the box for ionization descriptors checked on. The ionization descriptors should then be renamed as described above. Then the ADMET calculations should be performed again, this time with the box unchecked so that the pH dependent models will be calculated correctly by calculating the ionization descriptors at the default pH. In this case, additional columns will be generated for the ionization descriptors that had been renamed previously. For example, if after performing the first step in this process, the user renamed FCation as FCation@1.2, then in the second step, a new column will be generated with the name FCation corresponding to the value of FCation at the default pH of 7.4. The set of descriptors that depend on pH are: FAnion, FCation, FUnion, QAvgNeg, QAvgPos, F_NLP and F_HBP. Note that FZwitter is pH independent as it refers to the fraction of microspecies that are zwitterionic for the neutral (uncharged) macrostate, and this value does not depend on pH.
Note that both ionization descriptors and pH-dependent properties rely by default on pKa values calculated by the S+pKa model. Users can replace these predicted pKa with their own (e.g., experimental) supplied in the input file. See user pKa paragraph for more deatils.