CoChat
Sign Up
Cited by 45Drug Solubulity and Delivery SystemsComputational Drug Discovery MethodsPharmaceutical studies and practices

IMI – Oral biopharmaceutics tools project – Evaluation of bottom-up PBPK prediction success part 3: Identifying gaps in system parameters by analysing In Silico performance across different compound classes

Open access
Darwich, Adam S., Margolskee, Alison, Pepin, Xavier, Aarons, Leon, Galetin, Aleksandra, Rostami-Hodjegan, Amin, Carlert, Sara, Hammarberg, Maria +49 more·2017·European Journal of Pharmaceutical Sciences
DOI 10.1016/j.ejps.2016.09.037
Open this paper in CoChat — free accountOpen PDF

Abstract

&NA; Three Physiologically Based Pharmacokinetic software packages (GI‐Sim, Simcyp® Simulator, and GastroPlus™) were evaluated as part of the Innovative Medicine Initiative Oral Biopharmaceutics Tools project (OrBiTo) during a blinded “bottom‐up” anticipation of human pharmacokinetics. After data analysis of the predicted vs. measured pharmacokinetics parameters, it was found that oral bioavailability (Foral) was underpredicted for compounds with low permeability, suggesting improper estimates of intestinal surface area, colonic absorption and/or lack of intestinal transporter information. Foral was also underpredicted for acidic compounds, suggesting overestimation of impact of ionisation on permeation, lack of information on intestinal transporters, or underestimation of solubilisation of weak acids due to less than optimal intestinal model pH settings or underestimation of bile micelle contribution. Foral was overpredicted for weak bases, suggesting inadequate models for precipitation or lack of in vitro precipitation information to build informed models. Relative bioavailability was underpredicted for both high logP compounds as well as poorly water‐soluble compounds, suggesting inadequate models for solubility/dissolution, underperforming bile enhancement models and/or lack of biorelevant solubility measurements. These results indicate areas for improvement in model software, modelling approaches, and generation of applicable input data. However, caution is required when interpreting the impact of drug‐specific properties in this exercise, as the availability of input parameters was heterogeneous and highly variable, and the modellers generally used the data “as is” in this blinded bottom‐up prediction approach. Graphical Abstract Figure. No caption available.

AI snapshot

Written from the paper’s full text.

TL;DR

This study evaluated bottom-up PBPK predictions for 43 OrBiTo APIs across three software packages, focusing on how drug-specific properties affect prediction success. PBPK performed well for well-behaved compounds but showed gaps for acids, weak bases, low/high logP extremes, low BP/fu_p, and solubility/permeability-limited classes.

Key findings

  • For AUC0-t,last, neutral and strong bases performed well (80.0% and 92.3% within three-fold; AAFE 1.92 and 1.63), whereas acids performed poorly (10.0% and 20.0% within two- and three-fold; AAFE 6.33). Weak bases had poor F_oral predictions (25.0% within two-fold; AFE 1.48; AAFE 2.46).
  • Mid-range logP (Q2: 2.545–3.3) gave the best AUC0-t,last predictions (55.6% within two-fold; AFE 1.14; AAFE 2.17; R 0.862; CCC 0.836), while Q1 and Q4 had poor correlations (R 0.0411 and 0.0489). High logP Q4 underpredicted F_oral (AFE 0.477) and overpredicted Vd (AFE 2.43).
  • Higher BP quartiles improved predictions: AUC0-t,last within two-fold increased from 0.00% (Q1) to 63.6% (Q4), AAFE decreased from 5.91 to 2.09, and R increased from 0.0147 to 0.921; CL/CL/F predictions also improved with BP (R 0.197 to 0.793). Low fu_p tended to underpredict CL/CL/F.
  • BCS I had the best AUC correlation (R 0.850; CCC 0.837); BCS IV had lower predictability (33.33% within two-fold; 77.8% within ten-fold). F_oral for BCS III/IV showed negative bias (AFE 0.357) vs BCS I/II (AFE 1.54), and F_rel for BCS II/IV was underpredicted (AFE 0.677) vs BCS I/III (AFE 1.20).

Methods

  • Bottom-up PBPK simulations were generated for 43 APIs (from 83 in the OrBiTo database) using GastroPlus, Simcyp, and GI-Sim; participating institutions simulated allocated APIs, with overlap allowed to test user differences. Analysis focused on single-dose, fasted-state study arms only.
  • APIs were grouped by MW, acid/base nature, logP/logD, BP, fu_p, BCS class, dose number, and estimated f_a; continuous properties were split into quartiles. Simulated vs observed PK parameters included AUC0-t,last, F_oral, F_rel, Cmax, CL or CL/F, and Vd or Vd/F.
  • Performance metrics were % within two-, three-, and ten-fold, AFE, AAFE, Pearson R, and CCC; Pearson correlations among API properties were also calculated.

Limitations

  • Paper states: high degree of missingness and non-uniform API parameter data; lack of experimental protocol information; modellers were relatively unrestricted in input data selection and modelling approaches; simulations only for APIs meeting minimal data criteria; missing BP and logP replaced with estimates; without full input datasets model boundaries cannot be fully tested, and advanced models may collapse to simpler ones.
  • Paper states: interpretation is a function of data, model, and modeller; clinical data may be misrepresentative due to low sample sizes or high variability; models may fail to describe GI physiology, morphology, and processes governing F_oral.
  • Evident from text: several subgroup analyses are very small (e.g., BCS III n=2; D_o ≤1 F_rel n=3 APIs/86 simulations; logP Q2 F_oral n=1), and API properties are correlated (e.g., strong logP–fu_p correlation), which complicates attribution of performance gaps to single properties.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.1016/j.ejps.2016.09.037

Cites

14 of 14 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Claims the paper makes

0 of 11 cited sources checked · 0 supported · 0 flagged

  • Findingown data

    Examining predictive performance of simulated p.o. AUC0-t,last in relation to simulated average fup neglected to show any apparent trends.

    “Examining the predictive performance of simulated p.o. AUC 0-t,last in relation to simulated average fu p neglected to show any apparent trends.”
    Quote in the paperNot a substantial claim
  • Findingown data

    Great overlap between APIs of lowest BP quartiles and lowest fup group (fup < 0.01).

    “and a great overlap between APIs of the lowest BP quartiles and the lowest fu p group (fu p < 0.01).”
    Quote in the paperNot a substantial claim
  • Findingnone given

    BCS class II compound displayed similar level of percentage within two, three, ten-fold.

    “BCS class II compound displayed a similar level of percentage within two, three and ten-fold.”
    Quote in the paperNot a substantial claim
  • Findingnone given

    Predictions of p.o. AUC0-t,last performed poorly for APIs for which logP was not given (n=8) compared to those for which logP was reported (n=35).

    “Predictions of p.o. AUC 0-t,last performed poorly for APIs for which logP was not given (n=8) compared to those for which logP was reported (n=35).”
    Quote in the paperSubstantial claim
  • Findingnone given

    High BP tended to give better predictive performance of CL and CL/F.

    “High BP: Tended to give better predictive performance of CL and CL/F.”
    Quote in the paperNot a substantial claim
  • May combine two claims. 67
  • Classified as probably not a substantial claim. 15

  • May combine two claims.

    Findingown data

    More APIs in Q1 logD had predictions of F oral within two-fold compared with Q2-Q4 (66.7% vs 55.6%), and less within three and ten-fold, 66.7% and 66.7%, vs 88.9% and 100%, indicating results may be skewed by single outlier.

    “However, more APIs in Q1 had predictions of F oral within two-fold compared with Q2-Q4 (66.7% vs. 55.6%), and less within three and ten-fold, 66.7% and 66.7%, vs. 88.9% and 100%, for Q2-Q4, indicating the results may be skewed by a single outlier”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    Cmax best predicted for neutral compounds, strong and weak bases with 60.0%, 53.8%, and 45.5% within two-fold.

    “C max was best predicted for neutral compounds, strong and weak bases with 60.0%, 53.8% and 45.5% predicted with two-fold of observed data, respectively.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    Improvements seen in bias (AFE Q1-Q4: 0.247-0.75) and precision (AAFE Q1-Q4: 7.36-2.73) with increasing BP.

    “Similarly, improvements were seen in the bias (AFE Q1-Q4: 0.247-0.75)and precision (AAFE Q1-Q4: 7.36-2.73)with increasing BP”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    Q2 displayed largest bias (AFE 0.298, 3.36-fold underprediction) and poorest precision (AAFE 7.00), whereas Q1 lowest AFE 1.02 and Q4 lowest AAFE 2.05.

    “Q2 displayed the largest bias (AFE: 0.298, or a 3.36-fold underprediction) and poorest precision (AAFE: 7.00), whereas Q1 displayed the lowest AFE at 1.02 and Q4 the lowest AAFE at 2.05.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    Predictions for AUC0-t,last F oral poor for Q1 logD, AFE 0.418 (2.38-fold underprediction; n=3; vs AFE 1.14 for Q2-Q4; n=9), high variability AAFE 4.17 vs 1.87, poor correlation R -0.855, CCC -0.706.

    “Predictions for AUC 0-t,last F oral were poor for APIs in Q1 of logD, displaying AFE of 0.418 (or 2.38-fold underprediction; n=3; vs. AFE of 1.14 for Q2-Q4; n=9), high variability with AAFE of 4.17 (vs.1.87 for Q2-Q4), and poor correlation between predicted and observed with R of -0.855 and CCC of -0.706.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    Q1 logD showed high variability AAFE 7.60 vs 3.29, low correlation R -0.239, CCC -0.083.

    “Further, Q1 showed high variability with AAFE of 7.60 vs. 3.29 for Q2-Q4, and low correlation between predicted and observed with R of -0.239 and CCC of -0.083”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    fa ≥ 0.9 displayed lower bias, higher precision, and better correlation metrics for F oral predictions compared to fa < 0.9.

    “f a ≥ 0.9: Displayed a lower bias, higher precision and better correlation metrics for predictions of F oral as compared to the f a < 0.9 group.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    Q1 and Q2 for BP showed poor predictions of AUC0-t,last, low percentage within n-fold, high variability, poor correlations.

    “Q1 and Q2 for BP showed poor predictions of AUC 0-t,last , with a low percentage falling within n-fold, high variability, and poor correlations between predicted and observed”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    BCS IV APIs displayed lower degree of predictability with 33.33% within two-fold and 77.8% within tenfold; BCS class III limited to two APIs and could be considered undetermined.

    “BCS IV APIs displayed a lower degree of predictability with 33.33% falling within two-fold and 77.8% within tenfold; whereas analysis of BCS class III was limited to two APIs and could therefore be considered undetermined”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    Poor predictions of F oral for weak bases (25.0% within two-fold), tendency towards overprediction (AFE 1.48), high variability (AAFE 2.46), poor correlation R -0.56, CCC -0.296.

    “Poor predictions of F oral were seen for weak bases (25.0%within two-fold), showing a tendency towards overprediction (AFE 1.48), somewhat high source of variability (AAFE 2.46) and poor correlation between predicted and observed, with an R of -0.56 and CCC of -0.296”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    Investigating impact of BP on Cmax revealed overall trend of improvement towards compounds with higher BP, CCC ranged 0.224 for Q1 to 0.636 for Q4.

    “Investigating the impact of BP on the performance of predicting C max revealed an overall trend of improvement towards compounds with higher BP, for instance CCC ranged from 0.224 for Q1 to 0.636 for Q4.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    Similar trend for fup, Q1 and Q2 generally underperformed compared with Q3 and Q4 with low correlations.

    “A similar trend was seen for fu p , where Q1 and Q2 generally underperformed compared with Q3 and Q4 with low correlations between predicted and observed”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    Slight negative bias of F oral observed for acids with AFE 0.580 (1.72-fold underprediction) and poor correlation (R -0.570, CCC -0.480).

    “A slight negative bias of F oral was observed for predictions of acids with an AFE of 0.580 (1.72-fold underprediction) and poor correlation (R of -0.570 and CCC of -0.480).”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    Moderate positive correlations between logP and logD, logD and lowest acidic pKa, highest basic pKa and log(BP-0.5), and logit(fup) and log(BP-0.5).

    “Moderate positive correlations included logP vs logD, logD vs lowest acidic pKa, highest basic pKa vs log(BP -0.5), and logit(fu p ) vs log(BP -0.5).”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    Predictions for F rel between p.o. and solutions seemed to perform better for lower quartiles of logD, with 100% within two-fold for Q1, 75.0% for Q2 and Q3, 50.0% for Q4.

    “Predictions for F rel between p.o. and solutions seemed to perform better for lower quartiles of logD, with 100% within two-fold for Q1, 75.0% for each of Q2 and Q3, and 50.0%for Q4.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    Variability in predictions appeared to increase with increasing logD, AAFE 1.35, 1.70, 1.66, 1.88 for Q1-Q4.

    “The variability in the predictions appeared to increase with increasing logD, with AAFE of 1.35, 1.70, 1.66, and 1.88 for Q1, Q2, Q3 and Q4.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    Clearance predictions showed high variability and poor correlation for lower values of BP.

    “Clearance predictions also showed high variability and poor correlation for lower values of BP”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingreasoning

    Slight overprediction of F oral observed for weak bases, whereas F rel between p.o. formulations and solution was underpredicted.

    “A slight overprediction of F oral was observed, whereas F rel between p.o. formulations and solution was underpredicted”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    Relative AUC between p.o. formulations and solution predicted best for APIs in Q3 logP (3.3-4.49), with 100% within two-fold (n=4), AFE 1.06, AAFE 1.19, R 0.99, CCC 0.87.

    “Relative AUC between p.o. formulations and solution was predicted best for APIs in the third quartile of logP (Q3: 3.3-4.49),with 100% within two-fold (n=4), AFE of 1.06, low variability with AAFE of 1.19, and strong correlation with R of 0.99 and CCC of 0.87”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    Q3 and Q4 showed higher frequency within n-fold, 50.0, 58.3, 100% for Q3, 63.6, 81.8, 100% for Q4.

    “Q3 and Q4 showed higher frequency within n-fold, with 50.0, 58.3 and 100% for Q3, and 63.6, 81.8, and 100% for Q4 falling within two, three and ten-fold.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    APIs not reporting logD also had relatively unbiased estimates and reasonable variability.

    “APIs not reporting logD also had relatively unbiased estimates and reasonable variability compared with APIs reporting logD.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    Average predictive performance for Vd or Vd/F for i.v. and p.o. simulations relatively unbiased for low logP compounds with AFE 0.917 to 1.14 for Q1-Q3, while general trend of overprediction for highly lipophilic compounds (AFE 2.43 for Q4).

    “The average predictive performance for V d or V d /F for i.v. and p.o. simulations was relatively unbiased for low logP compounds with AFE ranging from 0.917 to 1.14 for Q1 through Q3, while a general trend of overprediction was observed for highly lipophilic compounds (AFE of 2.43 for Q4)”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    AUC0-t,last predictions for acidic compounds displayed low bias with AFE 1.16, indicating both over- and underpredictions, while i.v. formulations displayed overall overestimation.

    “The AUC 0-t,last predictions of p.o. simulations for acidic compounds displayed a low bias with AFE of 1.16 (Figure 2B), indicating the presence of both over-and underpredictions, while i.v.formulations displayed an overall overestimation as compared to observed data.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    Neutral compounds and weak bases displayed strongest correlation metrics (R and CCC of 0.783 and 0.644 for weak bases; R and CCC of 0.922 and 0.749 for neutral APIs).

    “Neutral compounds and weak bases further displayed the strongest correlation metrics (R and CCC of 0.783 and 0.644, for weak bases; R and CCC of 0.922 and 0.749 for neutral APIs”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    Both low logP (Q1) and high logP (Q4) performed poorly compared to Q2, with lack of correlations (R 0.0411 and 0.0489; CCC 0.0183 and 0.0481).

    “Both low logP (Q1: -0.72-2.54) and high logP (Q4: 4.49-7.75)APIs performed poorly compared to the Q2, displaying a lack of correlations, with an R of 0.0411 and 0.0489 for Q1 and Q4, respectively, and a CCC of 0.0183 and 0.0481”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    For BCS class I compounds, 37.5% of APIs displayed predicted average oral AUC0-t,last within two-fold, 87.5% within ten-fold.

    “For BCS class I compounds, 37.5% of APIs displayed a predicted average oral AUC 0-t,last within two-fold, with 87.5% of predictions falling within ten-fold”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    Q4 showed high correlations between predicted and observed, R 0.985, but relatively low CCC 0.510, suggesting predictions were in right direction but wrong scale.

    “Interestingly, Q4 showed high correlations between predicted and observed, with R of 0.985, but relatively low CCC of 0.510, suggesting the predictions were in the right direction but on the wrong scale”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    APIs in upper quartile of logP generally gave underpredictions of F oral (AFE 0.477 for Q4), while those in third quartile gave overpredictions (AFE 2.21).

    “APIs in the upper quartile of logP generally gave underpredictions of F oral (AFE of 0.477 for Q4), while those in the third quartile gave overpredictions (AFE of 2.21).”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    Q3 and Q4 displayed good correlation R 0.825 and 0.921, CCC 0.727 and 0.908.

    “In contrast, Q3 and Q4 displayed good correlation between predicted and observed data with R of 0.825 and 0.921 and CCC of 0.727 and 0.908”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    Q1 displayed 0.00, 18.2, 81.8% within two, three, ten-fold with similar Q2.

    “Q1 displayed 0.00, 18.2 and 81.8% of predictions within two, three and ten-fold with similar results for Q2.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    Poor predictions of AUC0-t,last for acids with 10.0% and 20.0% within two and three-fold for p.o. simulations and high variability AAFE 6.33.

    “Poor predictions of AUC 0-t,last were observed for acids with 10.0% and 20.0% within two and three-fold for p.o. simulations and a high variability with AAFE of 6.33”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    Percent within n-fold increased with increasing BP quartile, 0.00, 33.3, 50.0, 63.6% within two-fold for Q1-Q4, 27.3, 44.4, 66.7, 81.8% within three-fold, 72.7, 77.8, 100, 100% within ten-fold.

    “The percent within n-fold increased with increasing BP quartile, where 0.00, 33.3, 50.0 and 63.6% fell within two-fold for Q1 through Q4, respectively, 27.3, 44.4,66.7 and 81.8% within three-fold, and 72.7, 77.8, 100, and 100% within ten-fold.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    On average, bias for different logD quartiles appeared comparable, AFE 0.868, 0.829, 1.09, 0.606 for Q1-Q4.

    “On average, the bias for the different quartiles appeared comparable, with AFE of 0.868, 0.829, 1.09, and 0.606 for Q1, Q2, Q3, and Q4, respectively.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    Q1 showed relatively low bias AFE 0.951, Q2-Q4 continued trend of better performance with increasing BP, AFE 2.13, 1.57, 0.98.

    “While Q1 showed a relatively low bias with AFE of 0.951, Q2 through Q4 continued the trend of better performance with increasing BP, with AFE of 2.13, 1.57, and 0.98.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    However, increase in performance in terms of R and CCC with increasing percentiles, R -0.0658, 0.00985, 0.527, 0.866 for Q1-Q4, CCC -0.0614, 0.00851, 0.400, 0.810.

    “However, there was an increase in performance in terms of R and CCC with increasing percentiles, with R values of -0.0658, 0.00985, 0.527, and 0.866 for Q1 through Q4, and CCC of -0.0614, 0.00851, 0.400, and 0.810 for Q1 through Q4”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    AUC0-t,last overpredicted on average for Q1 logD, AFE 2.19 vs 1.15 for Q2-Q4.

    “AUC 0-t,last was overpredicted on average for APIs in Q1, with AFE of 2.19 vs. 1.15 for Q2-Q4 combined.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    Predictions for APIs in Q2 logP (2.545-3.3) showed tendency towards better performance vs extremes, with 55.6% within two-fold, AFE 1.14, AAFE 2.17, R 0.862, CCC 0.836.

    “Investigating the impact of logP on predictions of AUC 0-t,last revealed predictions for APIs in the second quartile range of logP (Q2: 2.545-3.3) to show a tendency towards better performance as compared to either extremes, with 55.6% within two-fold error, AFE of 1.14, AAFE of 2.17, and R of 0.862 and CCC of 0.836”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    Moderate negative correlations between highest basic pKa and log(Do), and between logP and log(BP-0.5).

    “Moderate negative correlations included highest basic pKa vs log(D o ) and logP vs log(BP -0.5).”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    Correlation coefficients R and CCC suggested similar correlations between predicted and observed F oral for two groups with slight favour towards fa ≥ 0.9 (R 0.517, CCC 0.439) vs fa < 0.9 (R 0.451, CCC 0.373).

    “Correlation coefficients R and CCC suggested similar correlations between predicted and observed F oral for the two groups with a slight favour towards f a ≥ 0.9 (R: 0.517, CCC: 0.439) as compared to f a < 0.9 (R: 0.451, CCC: 0.373”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    For F oral, compounds with estimated fa below 0.9 showed negative bias, with an AFE of 0.499 (n=6), compared with 1.47 for compounds with estimated fa at or above 0.9.

    “Similarly, F oral predictions for estimated f a < 0.9 displayed negative bias with AFE of 0.499 (n=6) vs. 1.47 for APIs with estimated f a ≥ 0.9 (Figure9E).”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    Compounds with Do > 1 (n=21) had an AFE of 0.681, compared with 1.38 for compounds with Do ≤ 1.

    “Similarly, compounds with D o > 1 (n=21) displayed AFE of 0.681 vs. 1.38 for compounds with D o ≤ 1 (Figure10E).”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    Apparent difference in R and CCC for predicted vs observed oral data for APIs divided based on Do cut-off, R 0.850 and 0.0684 for Do ≤1 and Do >1, CCC 0.837 and 0.0640.

    “There was an apparent difference in R and CCC for predicted vs. observed oral data for APIs divided based on the D o cut-off point, with calculated R coefficients of 0.850 and 0.0684 for D o ≤ 1 and D o > 1, , and CCC of 0.837 and 0.0640”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    Freely soluble APIs (Do ≤1) displayed slightly poorer performance in Cmax prediction with respect to percent within N-fold, AFE and AAFE, while displaying minor improvement in correlation between predicted and observed Cmax (R and CCC 0.627 and 0.533 vs 0.442 and 0.438 for solubility limited).

    “Freely soluble APIs (compounds with D o ≤ 1) displayed slightly poorer performance in C max prediction with respect to percent within N-fold, AFE and AAFE, while displaying a minor improvement in correlation between predicted and observed C max , (R and CCC of 0.627 and 0.533, respectively, vs. 0.442 and 0.438 for solubility limited compounds”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    fa ≥ 0.9 group displayed lower bias and higher precision compared to fa < 0.9 group, AFE 1.47 and 0.499, AAFE 1.83 and 2.53.

    “The f a ≥ 0.9 group displayed a lower bias and higher precision as compared to the f a < 0.9 group, with AFE of 1.47 and 0.499, respectively and AAFE of 1.83 and 2.53”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    One area of noticeable difference was overall bias of Frel predictions, Do >1 group displayed overall negative bias AFE 0.681 (1.47-fold underprediction) and Do ≤1 group displayed positive bias AFE 1.38.

    “One area of noticeable difference was in the overall bias of the F rel predictions, for which the D o > 1 group displayed an overall negative bias with AFE of 0.681 (a 1.47-fold underprediction) and the D o ≤ 1 group displayed a positive bias with an AFE of 1.38”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    Predictions of F oral in relation to estimated fa gave comparable percent within two and threefold whereas APIs with higher estimated fa ≥ 0.9 displayed larger percentage within tenfold compared to fa < 0.9 group (100% vs 83.3%).

    “Predictions of F oral in relation to estimated f a gave comparable percent within two and threefold whereas APIs with higher estimated f a ≥ 0.9 displayed a larger percentage within tenfold as compared to the f a < 0.9 group (100% vs. 83.3%”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    Analysis managed to highlight both cases where PBPK absorption modelling performed in line with clinical data and cases where simulations deviated.

    “The analysis of the simulation exercise managed to highlight both cases where the PBPK absorption modelling approach performed in line with clinical data and cases within the drug-specific parameter space where simulations deviated from the expected.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    There were only 86 simulations and 3 APIs in Do ≤1 group for which Frel was obtainable, whereas 594 simulations and 21 APIs in Do >1 enabling calculation of Frel.

    “There were only 86 simulations and 3 APIs in the D o ≤ 1 group for which F rel was obtainable, whereas there were 594 simulations and 21 APIs in the D o > 1 enabling calculation of F rel .”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    BCS class I compounds outperformed remaining classes, R 0.850, CCC 0.837.

    “With respect to correlation parameters, BCS class I compounds outperformed the remaining classes, displaying an R value of 0.850 and a CCC value of 0.837”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    Two groups displayed comparable percent within two, three, ten-fold (66.7%, 66.7%, 100% for Do ≤1 and 76.2%, 81%, 100% for Do >1), and comparable precision AAFE 1.66 and 1.67.

    “However, the two groups displayed comparable percent within two, three, and ten-fold (66.7%, 66.7% and 100%, respectively for D o ≤ 1 and 76.2%, 81% and 100% for D o > 1), and comparable precision with AAFE of 1.66 and 1.67 for D o ≤ 1 and D o > 1.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    For F oral, BCS class III and IV compounds (low permeability) showed negative bias relative to BCS class I and II (high permeability), with an AFE of 0.357 for classes III and IV combined (n=5) and 1.54 for classes I and II combined (n=12).

    “F oral predictions for BCS classes III and IV (low permeable compounds) displayed negative bias as compared with BCS I and II (high permeable compounds), displaying AFE of 0.357 for BCS III and IV combined (n=5) and 1.54 for BCS I and II combined (n=12; Figure8E).”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    Freely soluble APIs displayed 37.5, 62.5, 87.5% within two, three, ten-fold and solubility-limited compounds displayed 37.1, 54.3, 88.6%.

    “Freely soluble APIs displayed 37.5, 62.5 and 87.5% within two, three and ten-fold and solubility-limited compounds displayed 37.1, 54.3 and 88.6% within two, three and ten-fold, respectively.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    F rel predictions showed an overall negative bias, while the Do ≤ 1 group showed an overall positive bias.

    “Displayed an overall negative bias for predictions of F rel and the D o ≤ 1 group displayed an overall positive bias.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    The wide-ranging analysis in this exercise pinpointed multiple targets for later model enhancement and essential inputs for a reliable model, which a smaller evaluation of a data-rich compound set might not have revealed.

    “However, the broad spectrum analysis of this exercise was able to identify several areas for future model improvement and key inputs needed for building a robust model, which may not have been possible with a smaller scale evaluation of a more data-rich compound set (Table 3).”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    For F rel, BCS class II and IV compounds (low solubility) showed negative bias, with an AFE of 0.677 (a 1.48-fold underprediction), whereas BCS class I and III (high solubility) showed overprediction with an AFE of 1.20.

    “F rel predictions for BCS classes II and IV (low soluble compounds) displayed negative bias, with AFE of 0.677 (a 1.48-fold underprediction) as compared with 1.20 overprediction for BCS I and III (high soluble compounds) (Figure5H).”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Study design and conductnone given

    Limited standard operating procedures were provided for Simcyp and GI-Sim, and most decisions on parameter selection and simulation setup were left to individual modellers.

    “Limited standard operating procedures were provided for Simcyp and GI-Sim, however, most decisions on parameter data selection and simulation setup were left to the individual modellers at each institution.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim. May combine two claims.

    Study design and conductnone given

    The analysis focused on single dose and fasted state study arms only, excluding multiple dose and fed state simulations.

    “The analysis presented in this manuscript focused on single dose and fasted state study arms only (excluding: multiple dose and fed state simulations).”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Premisecitation

    Acid/base nature and lipophilicity influence dissolution, absorption, and disposition, and are used as PBPK input parameters to calculate solubility and account for prandial state.

    “acid/base nature, as well as lipophilicity, are thought to play important roles in dissolution, absorption, and disposition, and are often used as input parameters of the PBPK model to dynamically calculate solubility in the different segments of the GI tract and account for the influence of prandial state”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites Miller et al., 2011 source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Premisecitation

    Lipophilicity can affect dissolution in aqueous media, and highly lipophilic compounds can be subject to enhanced solubilisation by bile salts.

    “Lipophilicity may also affect dissolution in aqueous media, and highly lipophilic compounds can be subject to enhanced solubilisation by bile salts”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites Mithani et al., 1996 source

      Not checked in the free run.

  • May combine two claims.

    Premisecitation

    Acids and bases distribute differently into tissues depending on tissue composition and phospholipid affinity; in silico Vd predictions account for these differences.

    “Acids and bases have the potential to distribute differently into tissues in the body, depending upon the tissue composition and their affinity for different phospholipids; in silico predictions of volume of distribution account for tissue composition and these differences between acids and bases”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites Rodgers and Rowland 2006

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Interpretationreasoning

    This may be related to in silico methods utilised for predicting distribution into tissues, which varied between users and software.

    “This may be related to the in silico methods utilised for predicting distribution into tissues, which varied between users and software.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationown data

    Poor predictive performance for acids may be due to issues in predicting disposition, as F oral and F rel were generally well predicted (80% and 100% within two-fold).

    “This poor predictive performance may be due to issues in predicting disposition, as both F oral and F rel between p.o. formulations and solutions were generally well predicted (80% and 100% within two-fold)”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationown data

    Underpredictions of F rel for compounds with logP in Q4 suggest inadequate models for solubility and dissolution of highly lipophilic compounds, possibly underperforming bile enhancement models or lack of solubility data.

    “Underpredictions of F rel for compounds with logP values in Q4 suggest inadequate models for solubility and dissolution of highly lipophilic compounds, possibly underperforming bile enhancement models or lack of solubility data generated in biorelevant media for highly lipophilic compounds.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    High correlation between logP and plasma protein binding is well documented.

    “The high correlation between logP and plasma protein binding has also been well documented”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites Yamazaki and Kanaoka, 2004

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Interpretationreasoning

    Low BP indicates lack of uptake into red blood cells, which may reflect high plasma protein binding in these cases.

    “Low BP indicates lack of uptake into red blood cells, which may reflect the high plasma protein binding in these cases.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    Alternative models have been developed to account for overprediction in Vss for highly lipophilic compounds.

    “Alternative models have been developed to account for the overprediction in V ss for highly lipophilic compounds”
    Quote in the paperNot a substantial claim0 of 2 sources checked
    • cites Berezhkovskiy, 2004 source

      Not checked in the free run.

    • cites Poulin and Haddad, 2012

      Not checked in the free run.

  • May combine two claims.

    Premisecitation

    De Buck 2007 investigated prediction of hepatic clearance using two methods; method 2 performed considerably better at predicting in vivo hepatic clearance from in vitro CLint, potentially through unintentionally compensating for inherent underprediction in CLint when scaling from HLMs or human hepatocytes.

    “De Buck 2007 investigated the prediction of hepatic clearance using two different methods, method 1 using the traditional formula involving fu p /BP and fu inc , and method 2 where the effects of fu p /BP and fu inc were assumed to cancel each other out.The second method performed considerably better at predicting in vivo hepatic clearance from in vitro CL int , potentially through unintentionally”
    Quote in the paperSubstantial claim0 of 2 sources checked
    • cites De Buck 2007

      Not checked in the free run.

    • cites Hallifax et al., 2012 source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Interpretationreasoning

    Results of De Buck 2007 and relationship observed here between low BP and high variability, and low fup and high variability in prediction of clearance may be related.

    “The results of De Buck 2007 and the relationship we have observed here between low BP and high variability, and low fu p and high variability in the prediction of clearance may be related.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim.

    Study design and conductreasoning

    The approach can be considered an opportunistic blinded modelling exercise that was driven by the availability of parameter data.

    “As such, this approach therefore can be regarded as an opportunistic blinded modelling exercise driven by availability of parameter data.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S2 (checkable).

    Interpretationreasoning

    Under such complex conditions, modelling may depend more on the quality of input data, model assumptions, and the modeller's experience.

    “In such scenarios, modelling efforts may rely more heavily on quality of input data, model assumptions, and modeller experience.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Interpretationnone given

    Modellers' interpretation and selection of input parameters can significantly impact performance.

    “The modellers' interpretation and selection of input parameters can also significantly impact performance.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationreasoning

    For highly permeable compounds, fa saturates near 1.0, so it offers limited information about the model's predictive performance.

    “This is possibly not surprising, as for highly permeable compounds f a will saturate at a value close to1.0,thus providing limited information on prediction performance of the model.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S2 (checkable), S3 (load-bearing).

    Interpretationreasoning

    Performance is a function of data, model and modeller.

    “There are challenges in interpreting the results of this analysis, as performance is a function of data, model and modeller.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationown data

    The underprediction of F oral for low-permeability compounds might point to model oversensitivity to in vitro permeability, inaccurate intestinal surface area estimates, underestimated colonic absorption, or missing transporter data.

    “This underprediction of F oral for low permeable compounds could potentially indicate an oversensitivity of the models to in vitro permeability measurements, improper intestinal surface area estimates, underestimates of colonic absorption and/or lack of intestinal transporter information.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S4 (based).

    Interpretationnone given

    Models may fail to appropriately describe gastrointestinal physiology, morphology and underlying processes governing F oral.

    “Models may fail to appropriately describe gastrointestinal physiology, morphology and the underlying processes governing F oral .”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based). May combine two claims.

    Interpretationnone given

    Data to inform parameters may be varied quality and lacking, and clinical data may be misrepresentative due to low sample sizes or high variability.

    “Data to inform parameters may be of varied quality and in many cases was lacking, and clinical data may be misrepresentative due to low sample sizes or high variability.”
    Quote in the paperNot a substantial claim
76 of 76

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

  • Combining QSPR predictions with PBPK models from compound structure allows the assessment of risks based on compound type/properties, guides thinking and enables prioritisation of fit-for-purpose resources to support model development.

    “Combining QSPR predictions with PBPK models from compound structure allows the assessment of risks based on compound type/properties [91], guides thinking and enables prioritisation of fit-for-purpose resources to support model development.”

    From Physiologically Based Pharmacokinetic Modelling for First-In-Human Predictions: An Updated Model Building Strategy Illustrated with Challenging Industry Case Studies (2019) · cites this paper as [91]

    Section 2.1 groups compounds by drug-specific properties, and the Conclusion (page not available in our copy) reports that PBPK prediction performance declines as compound complexity increases. This supports assessing prediction uncertainty by compound properties. The passages do not establish the more specific claim that combining structure-derived QSPR predictions with PBPK models enables that assessment; the words after [91] are not attributed to this source. (LLM, conf=0.82)

    Experimental — accuracy not yet measured

Cites

Flagged first, then by citations

Cited by

Sorted by citations

Related

Authors
Darwich, A. S., Margolskee, A., Pepin, X., Aarons, L., Galetin, A., Rostami-Hodjegan, A., Carlert, S., Hammarberg, M., Hilgendorf, C., Johansson, P., Karlsson, E., Murphy, D., Tannergren, C., Thörn, H., Yasin, M., Mazuir, F., Nicolas, O., Ramusovic, S., Xu, C., Pathak, S. M., Korjamo, T., Laru, J., Malkki, J., Pappinen, S., Tuunainen, J., Dressman, J., Hansmann, S., Kostewicz, E., He, H., Heimbach, T., Wu, F., Hoft, C., Pang, Y., Bolger, M. B., Huehn, E., Lukacova, V., Mullin, J. M., Szeto, K. X., Costales, C., Lin, J., McAllister, M., Modi, S., Rotter, C., Varma, M., Wong, M., Mitra, A., Bevernage, J., Biewenga, J., Van Peer, A., Lloyd, R., Shardlow, C., Langguth, P., Mishenzon, I., Nguyen, M. A., Brown, J., Lennernäs, H., & Abrahamsson, B.
Affiliations
  • University of Manchester (GB)
  • AstraZeneca (United Kingdom) (GB)
  • AstraZeneca (Sweden) (SE)
  • Sanofi (France) (FR)
  • Sanofi (Germany) (DE)
  • Sanofi (United States) (US)
  • Simcyp (United Kingdom) (GB)
  • Orion Corporation (Finland) (FI)
  • Goethe University Frankfurt (DE)
  • Novartis (United States) (US)
  • AbbVie (Germany) (DE)
  • Simulations Plus (United States) (US)
  • Pfizer (United States) (US)
  • Pfizer (United Kingdom) (GB)
  • MSD (UK) Limited (United Kingdom) (GB)
  • Janssen (Belgium) (BE)
  • GlaxoSmithKline (United Kingdom) (GB)
  • Johannes Gutenberg University Mainz (DE)
  • Bristol-Myers Squibb (United Kingdom) (GB)
  • Uppsala University (SE)
Corresponding author
Margolskee, Alison
ORCID
Venue
European Journal of Pharmaceutical Sciences
Publisher
Elsevier BV
Published
2017-01
Volume
96
Pages
626-642
ISSN
0928-0987 (ISSN-L)
DOI
10.1016/j.ejps.2016.09.037
OpenAlex
W2525378244
Semantic Scholar
e027e51561759cbfe31b312201790ad11c4e1af9
Cited by
45
References
14
Citations by year
2025: 1 · 2024: 2 · 2023: 3 · 2022: 5 · 2021: 5 · 2020: 8 · 2019: 8 · 2018: 6 · 2017: 6 · 2016: 1
FWCI
3.97
Type
journal article
Language
en
Open access
Licence
Locations
  • European Journal of Pharmaceutical Sciences (publishedVersion)
  • PubMed (publishedVersion)
  • Research Explorer (The University of Manchester) (submittedVersion, other-oa)
Funders
Medical Research Council (MC_G1100157)
Topics
  • Life Sciences › Pharmacology, Toxicology and Pharmaceutics › Pharmaceutical Science › Drug Solubulity and Delivery Systems
  • Physical Sciences › Computer Science › Computational Theory and Mathematics › Computational Drug Discovery Methods
  • Health Sciences › Medicine › Pediatrics, Perinatology and Child Health › Pharmaceutical studies and practices
Keywords
physiologically based pharmacokinetic modeling, oral bioavailability, relative bioavailability, in silico prediction, GastroPlus, intestinal permeability, intestinal surface area, intestinal transporters, bile micelles, precipitation chemistry, weak acids, weak bases, poorly water-soluble compounds
MeSH terms
Administration, Oral, Biopharmaceutics, Computer Simulation, Drug Evaluation, Preclinical, Pharmaceutical Preparations, Forecasting, Humans, Intestinal Absorption, Models, Biological

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Claims read by deepseek-v4.1-flash under definition version 3 (extract-v3) on 2026-10-09.

Classified by jev-1.13-20260917 (gold-guideline-v0.2) on 2026-10-09.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review