Skip to main content

Bioavailability & Permeation

ChemAudit evaluates a molecule's potential for oral bioavailability and membrane permeation using two complementary models: the Bioavailability Radar (6-axis physicochemical profile) and the BOILED-Egg classification (GI absorption and BBB penetration).

Both models appear in the Scoring Profiles tab under the "Bioavailability & Permeation" card after validating a molecule.

Bioavailability Radar​

The Bioavailability Radar is a 6-axis normalized assessment of oral bioavailability. Each axis represents a physicochemical property critical for a drug to be absorbed orally. If all six properties fall within their optimal ranges, the molecule has an excellent oral bioavailability profile.

The Six Axes​

AxisPropertyOptimal RangeUnitWhy It Matters
LIPOWLOGP (Lipophilicity)-0.7 to 5.0—Governs membrane permeability. Too hydrophilic = poor absorption; too lipophilic = poor solubility and toxicity risk. Upper bound from Lipinski's Rule of Five.
SIZEMolecular Weight150 to 500DaLarger molecules diffuse poorly across membranes. Upper bound of 500 Da from Lipinski's Rule of Five.
POLARTPSA (Topological Polar Surface Area)20 to 130A²Reflects surface polarity. Values above ~140 A² correlate with poor intestinal absorption. Based on Veber's rules.
INSOLULogS (ESOL aqueous solubility)-6 to 0log mol/LA drug must dissolve in GI fluid before absorption. LogS below -6 indicates very poor solubility.
INSATUFraction Csp3 (sp3 carbon fraction)≥ 0.25ratioMeasures 3D character. Higher Fsp3 improves solubility and clinical success. From Lovering et al.'s "Escape from Flatland."
FLEXRotatable Bonds≤ 9countExcessive flexibility imposes entropic penalties on binding and reduces oral bioavailability. Based on Veber's rules.

Interpretation​

In-Range CountInterpretation
6/6Excellent bioavailability profile
4–5/6Good profile
2–3/6Moderate profile — some properties outside optimal ranges
0–1/6Poor profile — significant bioavailability concerns

Normalization​

Each axis is normalized to a 0–1 scale:

  • Within optimal range: normalized value = 1.0
  • Outside range: linearly decreases toward 0 based on distance from the range boundary, reaching 0 at a distance equal to the range width

Implementation Note​

ChemAudit uses WLOGP (Wildman-Crippen LogP via RDKit's Crippen.MolLogP) for the lipophilicity axis. The original SwissADME paper uses XLOGP3 for the radar. Both are well-validated lipophilicity descriptors with closely correlated values, and the same optimal ranges apply.

BOILED-Egg Classification​

BOILED-Egg stands for Brain Or IntestinaL EstimateD permeation. It is a graphical model that simultaneously predicts passive gastrointestinal (GI) absorption and blood-brain barrier (BBB) penetration using only two molecular descriptors:

  • X-axis: TPSA (Topological Polar Surface Area)
  • Y-axis: WLOGP (Wildman-Crippen LogP)

Classification Regions​

The model defines three zones on a TPSA vs WLOGP scatter plot:

RegionColorMeaningApproximate Boundaries
WhiteGreenHigh probability of passive GI absorptionTPSA < ~142 A², WLOGP between ~-2.3 and ~+6.8
YolkAmberHigh probability of BBB permeation (also GI absorbed)TPSA < ~79 A², WLOGP between ~+0.4 and ~+6.0
GreyGreyPredicted as neither passively GI-absorbed nor BBB-permeantOutside both ellipses

A molecule in the yolk is always also in the white — BBB-permeant compounds are predicted to be GI-absorbed as well, consistent with biological reality.

Elliptical Model​

Each region is defined by an ellipse in the TPSA-WLOGP plane. A molecule's (TPSA, WLOGP) point is tested against each ellipse using the standard point-in-ellipse equation:

((TPSA - cx) / a)² + ((WLOGP - cy) / b)² ≤ 1

Where (cx, cy) is the ellipse center and (a, b) are the semi-axes. The ellipse parameters were optimized using Monte-Carlo optimization evaluated by the Matthews Correlation Coefficient (MCC).

Accuracy​

MetricGI Absorption (White)BBB Permeation (Yolk)
Internal Accuracy93%90%
Internal MCC0.700.79
10-fold CV Accuracy92%88%
10-fold CV MCC0.650.75
External Validation (46 FDA NCEs)83%—

Training Data​

  • GI absorption dataset: 660 molecules (567 well-absorbed, 93 poorly absorbed)
  • BBB permeation dataset: 260 molecules (156 BBB-permeant, 104 non-permeant)

References​

  1. Bioavailability Radar / SwissADME: Daina A, Michielin O, Zoete V. SwissADME: a free web tool to evaluate pharmacokinetics, drug-likeness and medicinal chemistry friendliness of small molecules. Sci. Rep. 2017, 7, 42717. DOI: 10.1038/srep42717

  2. BOILED-Egg: Daina A, Zoete V. A BOILED-Egg to predict gastrointestinal absorption and brain penetration of small molecules. ChemMedChem 2016, 11(11), 1117–1121. DOI: 10.1002/cmdc.201600182

  3. Wildman-Crippen LogP (WLOGP): Wildman SA, Crippen GM. Prediction of physicochemical parameters by atomic contributions. J. Chem. Inf. Comput. Sci. 1999, 39(5), 868–873. DOI: 10.1021/ci990307l

  4. Lipinski's Rule of Five: Lipinski CA, Lombardo F, Dominy BW, Feeney PJ. Experimental and computational approaches to estimate solubility and permeability in drug discovery and development settings. Adv. Drug Deliv. Rev. 2001, 46(1-3), 3–26. DOI: 10.1016/S0169-409X(00)00129-0

  5. Veber Rules: Veber DF, Johnson SR, Cheng HY, Smith BR, Ward KW, Kopple KD. Molecular properties that influence the oral bioavailability of drug candidates. J. Med. Chem. 2002, 45(12), 2615–2623. DOI: 10.1021/jm020017n

  6. Escape from Flatland (Fsp3): Lovering F, Bikker J, Humblet C. Escape from flatland: increasing saturation as an approach to improving clinical success. J. Med. Chem. 2009, 52(21), 6752–6756. DOI: 10.1021/jm901241e

  7. ESOL Solubility: Delaney JS. ESOL: estimating aqueous solubility directly from molecular structure. J. Chem. Inf. Comput. Sci. 2004, 44(3), 1000–1005. DOI: 10.1021/ci034243x

Next Steps​