Theoretical Foundations

This document describes the scientific basis and algorithms underlying the admetshiny package.


Drug-Likeness Filters

Lipinski Rule of Five (1997)

Reference: Lipinski, C. A., Lombardo, F., Dominy, B. W., & Feeney, P. J. (1997). Experimental and computational approaches to estimate solubility and permeability in drug discovery and development settings. Advanced Drug Delivery Reviews, 23(1-3), 3-25.

Criteria:

Parameter Threshold
Molecular Weight (MW) <= 500 Da
LogP <= 5
H-bond Acceptors (HBA) <= 10
H-bond Donors (HBD) <= 5

Implementation: A compound violates if any criterion exceeds the threshold. Up to 1 violation is generally considered acceptable.

Ghose Filter (1999)

Reference: Ghose, A. K., Viswanadhan, V. N., & Wendoloski, J. J. (1999). A knowledge-based approach in designing combinatorial or medicinal chemistry libraries for drug discovery. Journal of Combinatorial Chemistry, 1(1), 55-68.

Criteria:

Parameter Range
Molecular Weight 160 - 480 Da
Molar Refractivity (MR) 40 - 130
LogP -0.4 to 5.6
Total Atoms 20 - 70

Note: The atom count refers to TOTAL atoms (heavy + hydrogens), consistent with the MW range. The filter prefers "#Total atoms" when available and falls back to "#Heavy atoms".

Veber Filter (2002)

Reference: Veber, D. F., Johnson, S. R., Cheng, H. Y., Smith, B. R., Ward, K. W., & Kopple, K. D. (2002). Molecular properties that influence the oral bioavailability of drug candidates. Journal of Medicinal Chemistry, 45(12), 2615-2623.

Criteria:

Parameter Threshold
Rotatable Bonds (RB) <= 10
TPSA OR HBA + HBD <= 140 OR <= 12

Note: The two polarity conditions are alternatives (OR), not cumulative, following the original paper.

Egan Filter (2000)

Reference: Egan, W. J., Merz, K. M., & Baldwin, J. J. (2000). Prediction of drug absorption using multivariate statistics. Journal of Medicinal Chemistry, 43(21), 3867-3877.

Criteria:

Parameter Threshold
TPSA <= 131.6
LogP <= 5.88

Muegge Filter (2001)

Reference: Muegge, I., Heald, S. L., & Brittelli, D. (2001). Simple selection criteria for drug-like chemical matter. Journal of Medicinal Chemistry, 44(12), 1841-1846.

Criteria:

Parameter Range/Threshold
MW 200 - 600
LogP -2 to 5
HBA <= 10
HBD <= 5
Rotatable Bonds <= 15
TPSA <= 150
Pharmacophore points >= 4

Note: Pharmacophore points are approximated as HBA + HBD (rings and hydrophobic centers are not directly available from CDK descriptors).


BOILED-Egg Model

Original Publication

Reference: Daina, A., & Zoete, V. (2016). A boiled egg to predict gastrointestinal absorption and brain penetration of small molecules. ChemMedChem, 11(11), 1117-1121.

WLOGP Polygons (Official)

The original BOILED-Egg was calibrated with WLOGP (Wildman-Crippen LogP). The official polygon coordinates are from the supplementary data (Data S3) of Daina & Zoete (2016). These define two elliptical regions in the TPSA-LogP space:

  • White region (yolk): Compounds predicted to be absorbed by the GI tract (HIA+)
  • Yellow region (white): Compounds predicted to cross the blood-brain barrier (BBB+)

ALogP Polygons (Trained)

When WLOGP is not available (e.g., CDK descriptors use ALogP), the package uses ALogP-trained polygons. These were trained using Monte Carlo optimization on experimental HIA and BBB data.

Training data: 659 HIA compounds and 256 BBB compounds from Daina & Zoete (2016).

Optimization method: Monte Carlo + Metropolis algorithm.

P-glycoprotein Substrate Prediction

Reference: Seelig, A. (1998). A general pattern for substrate recognition by P-glycoprotein. European Journal of Biochemistry, 251(1-2), 252-261.

The package uses an embedded Random Forest classifier trained from experimental data.

BOILED-Egg Training Section


P-glycoprotein Random Forest Model

Training Data

Source: Metrabase (J. Cheminformatics 2015, 7:21).

Compounds: 882 experimental compounds with P-gp substrate labels.

Model Architecture

Parameter Value
Algorithm Random Forest
Trees 100
Max depth 10
Features 9 CDK descriptors

Features

MW, LogP, TPSA, HBD, HBA, RB, HeavyAtoms, AromAtoms, MR

Performance

Metric Value
Cross-validated accuracy 69.4%
MCC 0.383

Statistical Methods

PERMANOVA (Anderson 2001)

Reference: Anderson, M. J. (2001). A new method for non-parametric multivariate analysis of variance. Austral Ecology, 26, 32-46.

Implementation: vegan::adonis2() on Euclidean distance of z-scored descriptors. 999 permutations.

Wilcoxon Rank-Sum Test (Mann-Whitney U)

Reference: Mann, H. B., & Whitney, D. R. (1947). On a test of whether one of two random variables is stochastically larger than the other. Annals of Mathematical Statistics, 18(1), 50-60.

Use: Comparing 2 groups (non-parametric alternative to t-test).

Kruskal-Wallis Test

Reference: Kruskal, W. H., & Wallis, W. A. (1952). Use of ranks in one-criterion variance analysis. Journal of the American Statistical Association, 47(260), 583-621.

Use: Comparing >= 3 groups (non-parametric alternative to ANOVA).

Effect size: eta-squared = H / (N - 1).

Post-hoc: Pairwise Wilcoxon with Holm Correction

Use: After significant Kruskal-Wallis (p < 0.05), identifies which specific pairs differ.

Correction: Holm-Bonferroni method for multiple comparisons.

Shapiro-Wilk Normality Test

Reference: Shapiro, S. S., & Wilk, M. B. (1965). An analysis of variance test for normality. Biometrika, 52(3-4), 591-611.

Use: Tests whether a variable follows a normal distribution.

Cophenetic Correlation

Reference: Sokal, R. R., & Rohlf, F. J. (1962). The comparison of dendrograms by objective methods. Taxon, 11(2), 33-40.

Use: Measures how faithfully a dendrogram preserves the original pairwise distances. Values close to 1.0 indicate a good fit.

Pearson Correlation

Reference: Pearson, K. (1895). Notes on regression and inheritance in the case of two parents. Proceedings of the Royal Society of London, 58, 240-242.

Spearman Rank Correlation

Reference: Spearman, C. (1904). The proof and measurement of association between two things. American Journal of Psychology, 15(1), 72-101.


Tanimoto Similarity

Reference: Willett, P., Barnard, J. M., & Downs, G. M. (1998). Chemical similarity searching. Journal of Chemical Information and Computer Sciences, 38(6), 983-996.

Implementation: Extended fingerprints computed via rcdk, Tanimoto similarity matrix, AGNES hierarchical clustering.

AGNES Linking Methods

Method Description
average (UPGMA) Average distance between clusters; balanced
complete Maximum distance (furthest neighbour); tight clusters
single Minimum distance (nearest neighbour); chaining effect
ward Minimizes within-cluster variance; compact spherical clusters

Additional Metrics

Pfizer 3/75 Rule

Reference: Hughes, J. D., et al. (2008). Estimating the toxicity of compounds. Bioorganic & Medicinal Chemistry Letters, 18(17), 4872-4875.

Criteria: LogP < 3 AND TPSA > 75.

GSK 4/400 Rule

Reference: Gleeson, M. P., et al. (2011). Drug-likeness. Journal of Medicinal Chemistry, 54(13), 4459-4468.

Criteria: MW <= 400 AND LogP <= 4.

Golden Triangle

Reference: Johnson, T. W., et al. (2009). Using the Golden Triangle. Bioorganic & Medicinal Chemistry Letters, 19(17), 4872-4875.

Criteria: MW 200-450 AND LogP -2 to 4.5 (oral bioavailability zone).

Lead-Likeness

Reference: Teague, S. J., et al. (1999). The design of leadlike combinatorial libraries. Angewandte Chemie International Edition, 38(24), 3743-3748.

Criteria: MW <= 450 AND LogP <= 4.5 AND rotatable bonds <= 10.


Model Training

Back to top