Skip to main content
. 2025 Jul 31;17(8):1002. doi: 10.3390/pharmaceutics17081002

Table 4.

Performance of computational methods in predicting the pharmacokinetic properties of natural compounds compared to experimental data.

Method Example Experimental Validation Performance Summary
Quantum Mechanics (QM) Stereoselectivity of nicotine hydroxylation by CYP2A6 [277] Yes (retrospective), computed (~97%) vs. wet lab (89–94%) High agreement
LogP estimation for BBB permeability (e.g., caffeine) [278] Yes (clinical data, retrospective) Good match with known BBB-crossing compounds
DFT used for C-H bond energy at the main metabolic site (e.g., acetic acid) [279] Yes (vs. experimentally derived bond energy) Lower bond dissociation energy at main metabolic site confirmed by experimental data as compared with other C-H bond
Global reactivity of 4-hydroxyisoleucine [280,281] Indirect (predicted stability vs. independent plasma stability study) Supported by external experimental data
Molecular Docking Docking of drugs with CYP2D6 variants [280] Yes, retrospective correlation (R2 = 0.81–0.92) High agreement
Flavonoids binding to Pgp [281] Very weak; correlation r = –0.27 to 0.079 Poor correlation despite otherwise claims
Lignans and flavonoids binding to Pgp [51] Partial; 2 of 10 flavonoids experimentally confirmed Partial success (at least 20%)
Abietane diterpenes binding to Pgp [282] Yes, for 2 hemisynthesis compounds Good performance for two tested compounds
Pharmacophore Models URAT1 inhibitors [105] Yes, 3 flavonoids of 25 hits were active (relatively low potency) Modest performance
CYP2D6 inhibitors [110] Yes; 42% strong, 33% moderate inhibition High agreement (75% activity in vitro)
DDIs via CYP1A2, 2C9, and 3A4 enzymes [283] Yes (vs in vitro results obtained with fluorescence-based P450 microarrays) 32.1–65.5% depending on model and enzyme
CYP3A4 inhibitors from Tripterygium wilfordii [284] Ye (vs. in vitro enzyme inhibition assays); 3 of 5 predicted were confirmed Good agreement
CYP1A2 inhibitors from herbal compounds [285] Yes; 7 of 12 compounds active ~58% accuracy for a combined approach (docking + pharmacophore models)
QSAR Models COMFA/COMSIA for natural phenolics [163] Yes; retrospective (r2pred = 0.78, 0.70) Very good agreement
Intestinal absorption prediction [286] Yes; 83% predictions within 2-fold of observed values Comparable to in vitro method
Drug absorption in rats [287] Reliability comparable to the Caco-2 and 2/4/A1 cell lines Very good agreement
Molecular Dynamics (MD) Withaferin-A and withanone membrane permeability [288] Yes; imaging based on antibody detection confirmed MD predictions Excellent agreement
Curcumin and quercetin binding to CYP3A4 and displacing CDK inhibitors [289] Yes; docking, MD, and IC50 (in vitro) Excellent agreement in several validation approaches
PBPK Models Oxymatrine dose prediction [290] Yes; compared to clinical dose Predicted dose (367 mg TID) aligned with clinical recommendation
Prediction of DDIs for hyperforin with sedative-hypnotics in human patients [291] Yes—model predictions compared with known clinical interactions Close agreement, all predictions within acceptable margin of error
PK of hydrastine and berberine [292] Yes—validated against observed clinical data Close fit to human PK data
PK of single dose and multiple dose administration of piperine [293] Yes—validated against actual clinical data All error values below the two-fold acceptance criterion