The CYP_Risk model is comprised of eight rules, each with a weight of 1, for CYP substrates only:
|
1A2: |
CYP1A2_CLint > [20,40] |
|---|---|
|
2C9: |
CYP2C9_CLint > [10,20] |
|
2C19: |
CYP2C19_CLint > [20,40] |
|
2D6: |
CYP2D6_CLint > [10,20] |
|
3A4: |
CYP3A4_CLint > [20,50] AND CYP3A4_HLM_CLint > [30,75] |
|
CL: |
CYP_HLM_CLint > [200,300] OR HEP_hCLint > [60,100] |
In the above, Km predictions are in μM, while intrinsic clearances are scaled to the respective CYP expression levels expected for human liver microsomes (HLMs), i.e., μL/min/mg HLM protein. Brackets indicate the range spanned by soft thresholds. See the section on ADMET Risk for further details.
The ADMET Risk thresholds represent a calibration of our models and are not directly based on experimental results. It is gratifying to note, then, that our generic CYP thresholds of around 30 μL/min/mg HLM protein are close to that cited for overall CYP clearance of antimalarial drugs (Gujjar 2011). Pfizer allows its CNS candidates to have higher clearance rates - about 100 μL/min/mg microsomal protein (Meanwell 2011). The fraction of substrates flagged by this threshold is 13% for CYP3A4_CLint and 22% for CYP3A4_HLM_CLint.
The fraction of the focused WDI that exceeds the upper CLint thresholds for CYPs 1A2, 2C9, 2C19 and 2D6 range from 8 to 11%. A greater proportion (19%) exceeds the upper threshold for recombinant CYP 3A4, in part because some of the WDI compounds are prodrugs. CYP3A4_CLint and CYP3A4_HLM_CLint criteria were combined to address this and to offset the effect of differences in P450 reductase: CYP ratios and cytochrome b5 content between the recombinant and microsomal assay systems (Rodrigues 1997). The top end of the resulting 3A rule “flags” 12% of the focused WDI.
CYP_Risk is greater than 2.0 for 9% of the compounds in the focused WDI after default out-of-scope penalties have been factored in.
Technical note
The training sets from which clearance rates are calculated necessarily consist only of substrates. Hence the CYP1A2_CLint prediction, for example, is effectively set to 0 if the associated CYP1A2_Substr classification is No by making the associated CYP***_Substr model the determining descriptor for each CYP***_CLint model. A side effect of this arrangement is that the CYP1A2 rule actually found in the ADMET_Risk and CYP_Risk model files is:
1A2: (CYP1A2_CLint > [20,40] OR ( CYP1A2_Substr = No AND CYP1A2_Substr = Yes )
The seemingly paradoxical AND clause contained within the parentheses is needed to produce the desired behavior when a CYP1A2_Substr prediction goes out-of-scope. Please see the Out-of-Scope Factor section of the Manual for a more detailed discussion.