Research Articles
Quantitative structure−pharmacokinetic parameters relationships (QSPKR) analysis of antimicrobial agents in humans using simulated annealing k‐nearest‐neighbor and partial least‐square analysis methods*

https://doi.org/10.1002/jps.20117Get rights and content

Abstract

We have developed quantitative structure−pharmacokinetic parameters relationship (QSPKR) models using k‐nearest‐neighbor (k‐NN) and partial least‐square (PLS) methods to predict the volume of distribution at steady state (Vss) and clearance (CL) of 44 antimicrobial agents in humans. The performance of QSPKR was determined by the values of the internal leave‐one‐out, crossvalidated coefficient of determination q2 for the training set and external predictive r2 for the test set. The best simulated annealing (SA)‐kNN model was highly predictive for Vss and provided q2 and r2 values of 0.93 and 0.80, respectively. For all compounds, the model produced average fold error values for Vss of 1.00 and for 93% of the compounds provided predictions that were within a twofold error of actual values. The best SA‐kNN model for prediction of CL yielded q2 and r2 values of 0.77 and 0.94, respectively, and had an average fold rror of 1.05. Use of PLS methods resulted in inferior QSPKR models. The SA‐kNN QSPKR approach has utility in drug discovery and development in the identification of compounds that possess appropriate pharmacokinetic characteristics in humans, and will assist in the selection of a suitable starting dose for Phase I, first‐time‐in‐man studies.

Section snippets

INTRODUCTION

During the drug development process, prediction of human pharmacokinetic parameters such as clearance and volume of distribution is fundamental in the design, optimization, and selection of lead drug candidates and in the determination of optimal dosing regimens for early phase clinical trials. In recent years, the application of combinatorial chemistry methods in drug discovery has produced vast numbers of drug candidates, which have dramatically increased the demand for rapid and efficient

Data

Pharmacokinetic data in humans for clearance (CL) and volume of distribution at steady state (Vss) were obtained from published literature for 44 antibiotics from different structural and therapeutic classes and are listed in Table 1.17 Parameters used in this analysis were derived from data following iv administration.

Generation of Molecular Descriptors

All chemical structures were generated using Sybyl version 6.7 (Tripos Associate, St. Louis, MO). Molecular Operating Environment (MOE) descriptors were generated for variable

Data and Molecular Descriptors

The 44 antibiotics included in the complete data set consisted of 18 cephalosporins (40.9%), nine penicillins (20.5%), four aminoglycosides (9.1%), two tetracyclines (4.5%), one macrolide (2.3%), one quinolone (2.3%), and nine antibiotics from nine other classes (20.4%) (Table 1). Scaled data were used for model building (standardized by mean and standard deviation). One hundred eighty‐one MOE descriptors were determined for each drug in the original data set using the QuaSAR‐descriptor option

DISCUSSION

In this study, we used a recently developed, variable‐selection kNN method18 to obtain QSPKR models for 44 antimicrobial agents. We compared these results to those obtained with models developed with the more commonly used PLS method. Using a combination of 2D and 3D MOE descriptors of chemical structures, we built a QSPKR model with high values of q2 (for the training and complete sets) and r2 (for the test set) using the SA‐kNN method. Efforts to build similar quality models with the PLS

Acknowledgements

This paper was presented in part at the Annual Meeting of the American Association of Pharmaceutical Scientists held in Toronto in 2003.

REFERENCES (21)

There are more references available in the full text version of this article.

Cited by (41)

  • Evaluating the influence of half-life, milk:plasma partition coefficient, and volume of distribution on lactational exposure to chemicals in children

    2017, Environment International
    Citation Excerpt :

    Influential parameters include lipophilicity, molecular weight, and heteroaromatic ring count and acidity. QSPR models have also been developed to predict chemical biological half-life or clearance in humans, with R2 and Q2 values ranging from 0.64 to 0.94 and 0.57 to 0.77, respectively (Karalis et al., 2003; Ng et al., 2004; Yap et al., 2006; Yu, 2010). Lipophilicity and hydrogen binding properties have been identified as influential parameters.

  • Chemometrics tools in QSAR/QSPR studies: A historical perspective

    2015, Chemometrics and Intelligent Laboratory Systems
    Citation Excerpt :

    Then, Sutter et al. in 1995 utilized GSA by some corrections for descriptor selection in QSAR [111]. After that, simulated annealing was employed in several QSAR studies for descriptor selection purpose in combination with various linear/non-linear mathematical methods, e.g. SA-MLR [232,233], SA-PCR [233], SA-PLS [234–238], SA-kNN (k-nearest-neighbor) [239,240] and SA-ANN [65,241]. SA has been also combined with GA and ANN to obtain a two-stage evolutionary algorithm for variable selection which showed good ability in non-linear model construction using radial basis function neural network [242].

  • Prediction of steady-state volume of distribution of acidic drugs by quantitative structure-pharmacokinetics relationships

    2012, Journal of Pharmaceutical Sciences
    Citation Excerpt :

    Several studies have been published proving the utility of QSPkR models for the prediction of VD. Most of them focused on congeneric series of drugs,5–10 and their predictive power is restricted within the respective structural class. Generation of adequate QSPkR models for heterogeneous series of structurally diverse drugs is a rather complicated task.

  • Prediction of volume of distribution values in human using immobilized artificial membrane partitioning coefficients, the fraction of compound ionized and plasma protein binding data

    2009, European Journal of Medicinal Chemistry
    Citation Excerpt :

    It is an important parameter to calculate the mean residence time MRT (MRT = VDss/Cl) and half-life of a drug (t1/2 = 0.693(VDss/Cl)), indicative of the drug exposure profile and the dosing frequency for setting up a suitable dosing regimen, which is intrinsically linked to other factors such as the pharmacodynamics and the therapeutical index. Various methods have been proposed to characterize VD including (i) the extrapolation of animal data [3–8], (ii) physiologically based pharmacokinetic (PBPK) model [6,9–13] and (iii) computational (in silico) approaches [14–20], Additionally, two biomimetic-binding measurement, the phospholipids binding (immobilized artificial membrane IAM) and the plasma protein binding (human serum albumin HSA) have been used to model VD in human [21]. These methods have been discussed in detail in several review papers [22–25].

View all citing articles on Scopus
*

This paper was presented in part at the Annual Meeting of the American Association of Pharmaceutical Scientists in Toronto in 2003.

View full text