Predicting Drug-plasma Protein Binding From Molecular Structure Information

Most drugs can bind with plasma proteins (serum albumin, α 1-acid glycoprotein and lipoproteins) to form reversible drug-plasma protein complex when drugs are absorbed in the blood. The complex can not penetrate through many biomembranes such as capillary vessel, blood-brain barrier and glomerulonephritis. So the bound drugs can not exhibit pharmacological activity and can not be metabolized and eliminated. But the reversible complex in plasma can serve as drug reservoir, replenish the free concentration of the drug in vivo when the elimination process has depleted the concentration of a free drug. So the extent to which the drugs bind to plasma proteins is greatly correlated with the pharmacokinetics properties (distribution, metabolism and excretion), the pharrnacodynamic and toxicity properties of the drugs. It is an important activity for pharmacokinetics and clinic pharmacology to study drug-plasma protein binding. Studying drug-plasma protein binding is usually through experiments. But the disadvantage of experiments is time-and money-consuming. Therefore more and more researchers try to predict drug-plasma protein binding from molecular structure information.In this paper, the models of predicting some types of drugs such as cephalosporins, quinolones, Beta-adrenoceptor blocking agents and xanthones binding with human serum albumin have been studied. Also the model of predicting some heterogeneous set of compounds binding with human serum albumin and some drugs of different molecular structure binding with bovine serum albumin using molecular structure parameters have been studied. The aim is to provide the method of predicting drug-plasma protein binding for high-through put screening new drugs.The major interactions of drug-plasma protein binding includ hydrogen bond, hydrophobic bond and ionic bond. In this paper, the theoretical structure parameterssuch as molecular weight (MW), molecular volume (V), the polar molecular surface area (PSA), indicator variable of acid and base (I) and dissociation constant (pka) which are greatly correlated with the interaction of drug-plasma protein binding have been used for establishing the model of predicting drug-plasma protein binding. The polar molecular surface area (PSA) is divided into the surface area of hydrogen bond donor (Sh) and the surface area of hydrogen bond acceptor (So,n) according to their different functions of hydrogen bond-forming. In this paper, Mont Carlo method was used to obtain V, Sh and So,n from their minimum energy conformations obtained from the optimization of the standard molecular geometry with the semiempirical self-consistent field molecular orbital calculation AMI method. The stepwise multiple regression analysis and the artificial neural network were used to establish the models of drug-plasma protein binding from molecular structure parameters.The regression equations and the BP neural networks of the fraction of 22 cephalosporins and 28 quinolones bound to human serum albumin with the surface area of hydrogen bond donor (Sh) and molecular weight are established respectively;the regression equation of the fraction of 13 Beta-adrenoceptor blocking agents bound to human serum albumin with the surface area of hydrogen bond donor (Sh), the surface area of hydrogen bond acceptor (So,n) and molecular weight is established;the regression equation and the BP neural network of the ability of 26 xanthones bound to human serum albumin with the surface area of hydrogen bond donor (Sh) and molecular volume are established;the regression equation and the BP neural network of the fraction of 19 homogeneous set of compounds bound to human serum albumin with the surface area of hydrogen bond acceptor (So,n) and dissociation constant (pka) are established;the regression equation of the fraction of 13 compounds bound to bovine serum albumin with indicator variable of acid and base (I) and molecular volume (V) is established. And in this paper, The BP neural network is established to predict the fraction of three types of drugs including 22 cephalosporins, 28 quinolones, and 13 Beta-adrenoceptor blocking agents bound to human serum albumin (neural units of the output layer) when the neural units of the input layer are the surface area of hydrogen bond donor (Sh), the surface area of hydrogen bondacceptor (So,n) and dissociation constant (pka). These models have many advantages such as simple calculation, clear physical meaning and good predictive results.According to the results of studies, drug-plasma protein binding is greatly correlated with molecular structure parameters such as molecular volume, molecular weight, the polar molecular surface area and dissociation constant. Predicting drug-plasma protein binding from molecular structure parameters which can be calculated easily is convenient and fast and it can be used to study the correlative parameters of pharmacokinetics.