Oral Bioavailability

Overview

This module predicts human oral bioavailability for new compounds using a combination of two methods - probabilistic and "mechanistic". First, the "mechanistic" part of the module tries to evaluate the query compound with regard to several crucial bioavailability limiting factors: solubility in stomach or intestine, stability in acidic media, intestinal membrane permeability by passive or active transport, P-gp efflux, and first pass metabolism in liver. All of these properties are predicted using independent algorithms, and experimental data sets. Results of the "mechanistic" evaluation of the human oral bioavailability are presented as easily interpretable "traffic-lights". In addition, this color-coded profile is converted to the probabilities for the compound's oral bioavailability to exceed 30% and 70% respectively with the help of multiple Recursive Partitioning trees. The final compound classification with regards to its oral bioavailability at clinically relevant doses is based on the cumulative result of those two statistical models.

Features

• ...

Interface

File:oral bioavailability.png

1. Dose...
   File:oral bioavailability simulation.png
   
   a. Click the "Undo" button to restore the automatically calculated property value for a compound and recalculate %F using default parameter values
   :b. Click to recalculate the %F using the currently specified dose
2. Classification of compounds based on predicted oral bioavailability (poor <30%, moderate 30-70%, good >70%). Probabilities that %F is greater than standard cut-off values are also presented.
3. Click to see more details regarding the calculation of the particular property
4. Factors affecting oral bioavailability (see below for details)
5. Hover over a title to view a screentip with a short description
6. Up to 5 similar structures in the %F (oral) DB with experimental values and references



Traffic-lights system explanation

1. Solubility in unbuffered water:
   • Green – good solubility in unbuffered water, Log S_w > -4 for electrolytes and Log S_w > -3 for non-electrolytes.
   • Red – very poor solubility in unbuffered water (25°C), Log S_w < -6 for electrolytes and Log S_w < -4.5 for non-electrolytes.
   • Yellow – moderate solubility.
2. Stability – susceptibility to acid hydrolysis in stomach:
   • Red – only assigned to highly reactive compounds that decompose in stomach very quickly. Red light means that F (oral) <10%, overriding any probabilistic prediction (i.e., do not pay attention to predicted probabilities in this particular case).
3. Passive absorption – ability to cross human intestinal membrane by passive diffusion:
   • Red – intestinal passive absorption <30%. %F (oral) is always less than passive absorption.
   • Green – good (>70%) passive absorption across intestinal barrier. Passive absorption does not effect %F.
4. First-pass metabolism – susceptibility to metabolic transformations catalyzed by enzymes in liver and intestine:
   • Red – high probability that first-pass metabolism is >50%. In this case %F (oral) is likely to be dose dependent and not to exceed 40%.
   • Green – compound probably does not undergo significant first-pass metabolism.
5. P-gp efflux – susceptibility to backward transport through intestinal membrane:
   • Red – compound is P-glycoprotein substrate. This effect is mostly important when compound is metabolized by CYP3A4
6. Active transport – susceptibility to active transport through intestinal membrane:
   • Green – compound is actively transported by PepT1, ASBT or other enzymes. Red light never appears, as this factor can only increase %F (oral).








Technical information




Training set size: 788
Internal validation set size: N/A*
* - Since the formal basis for the probabilistic prediction still are "mechanistic" considerations, there was no particular emphasis on external model validation - similarly to the knowledge based "expert-like". As result, the statistical results obtained for the training set compounds are the only available quantitative characteristics for the ACD/Oral Bioavailability model.

Main sources of experimental data:

• Reference books:
  • Therapeutic Drugs, Dolery, C., Ed. 2nd Edition, Churchill Livingstone, New York, NY, 1999
  • Clarke's Isolation and Identification of Drugs, Moffat, A.C., Jackson, J.V., Moss, M.S., Widdop, B., Eds. 2nd Edition, The Pharmaceutical Press, London, 1986
• Various articles from peer-reviewed scientific journals*

* - Both articles reporting oral bioavailability models by other authors (i.e. with larger collections of experimental data per article) and dealing with detailed experimental pharmacokinetic characterization (i.e. usually only several compounds per article) were available for the training set construction.







Training set statistics




Table 1. Classification performance of the lower threshold (%F (oral) > 30%) probabilistic bioavailability model on the training set compounds.

Subset	Observed	Calculated probability (p)
		>0.5	<0.5
Entire training set N = 788*	True	415 (52.7%)	114 (14.5%)
	False	81 (10.3%)	178 (22.6%)
	Accuracy	75.2%
	Sensitivity	78.4%
	Specificity	68.7%

* - Since ACD/Bioavailability model does not utilize GALAS modeling methodology, no additional parameters (i.e., RI) are available to enable additional filtering of more reliable predictions.




Table 2. Classification performance of the upper threshold (%F (oral) > 70%) probabilistic bioavailability model on the training set compounds.

Subset	Observed	Calculated probability (p)
		>0.5	<0.5
Entire training set N = 788*	True	172 (21.8%)	126 (16.0%)
	False	44 (5.6%)	446 (56.6%)
	Accuracy	78.4%
	Sensitivity	57.7%
	Specificity	91.0%

* - Since ACD/Bioavailability model does not utilize GALAS modeling methodology, no additional parameters (i.e., RI) are available to enable additional filtering of more reliable predictions.