23-06-2014, 11:41 AM
Prediction of the Penetration of Drugs by Artificial Neural Networks
[attachment=65192]
Abstract
In this study an Artificial Neural Network was
developed to predict the penetration of drugs through a
polydimethylsiloxane membrane by molecular descriptors. A
total of 245 drugs and their absorption experimentally
determined values were arranged into various data sets to
perform training and validation of the different implemented
neural networks. Logarithms of the maximum steady-state flux
(log J) values were correlated with four input variables; i) Count
fo H-Acceptor Sites (CHA), ii) H-Donors Charged Surface Area
(HDCA), iii) Gravitational index (Gb) and iv) Weighted Positive
Charged Partial Surface Area (WPSA-2). Besides these, other
neural networks were implemented with an extra input variable,
which measuring the similarity of the different drugs (Distance).
All developed neural networks present a high squared
correlation coefficient with a low root-mean-square error, and
they improve the MLR prediction model in 24%.
NTRODUCTION
A drug to be introduced into the body has to overcome
numerous barriers. For this, it undergoes through an absorption
process through several semipermeable biological membranes
to finally reach the blood. Bloodstream distributes the drug
throughout the body, where it is metabolized or eliminated by
excretion [1]. Therefore, absorption is a pharmacokinetic
process of a molecule entrance into the body to subsequently
reach systemic circulation. Biological membranes act as a filter
of different compounds which penetrate into cells. Thus,
understanding the mechanisms of drugs transport across
biological membranes, as skin, becomes necessary [2].
The main function of human skin is to protect the body
from foreign substances. There have been several studies to
increase transdermal absorption of drugs, but human skin
obtainment represents an obstacle. Synthetic polymer
membranes, such as Polydimethylsiloxane (PDMS), have been
widely employed as a substitute of human skin by many
researchers in the evaluation of transdermal drug delivery
systems [3]. Mouse skin has been widely employed in
various drug release studies. Those studies have shown that
diffusion rates of PDMS membranes are related to the
absorption through the skin [4,5]. Therefore, PDMS
membranes provide accurate and viable
MATHERIAL AND METHODS
Polar compounds pass slower than non-polar compounds
through PDMS membranes. Moreover, ionic compounds do
not pass through the membrane [5]. If diffusion is membrane
controlled and diffusant concentration in the solvent is
maintained at a negligible level, steady-state flux could be
calculated. On the other hand, steady-state flux reaches a
maximum value when the donor concentration is kept at the
solubility limit [5]. In this case, membrane permeability and
diffusant solubility values are comparable. Nevertheless,
permeability value is complicated to calculate. Furthermore, a
drug flux across the membrane is difficult to determinate and
parameter values are also complicated to obtain [5,8].
Previously mentioned inconvenient could be avoided by
using Artificial Neural Networks (ANNs) models, which are
capable of solving highly nonlinear problems. ANNs have been
extensively applied for solving several problems in various
areas [9-12], providing successfully results on complex
problems.
CONCLUSIONS
Different models have been developed with four and five
neurons in the input layer (CHA, Gb, HDCA, WPSA-2 and
Distance). All models have developed a better fit than the MLR
model developed by Weiping et al. [5] improving the model
between 13.5% and 33.02%, in Global Error terms.
The predictive power decreases, clearly with the number of
validation cases, going from 26% for models with 0%
validation cases to 14%-23% for models with 33% validation
cases. All models show a good fit in terms of RMSE and %
Error for training and validation phases. The results indicate
that neural networks have a good predictive capacity for log J
values.
[attachment=65192]
Abstract
In this study an Artificial Neural Network was
developed to predict the penetration of drugs through a
polydimethylsiloxane membrane by molecular descriptors. A
total of 245 drugs and their absorption experimentally
determined values were arranged into various data sets to
perform training and validation of the different implemented
neural networks. Logarithms of the maximum steady-state flux
(log J) values were correlated with four input variables; i) Count
fo H-Acceptor Sites (CHA), ii) H-Donors Charged Surface Area
(HDCA), iii) Gravitational index (Gb) and iv) Weighted Positive
Charged Partial Surface Area (WPSA-2). Besides these, other
neural networks were implemented with an extra input variable,
which measuring the similarity of the different drugs (Distance).
All developed neural networks present a high squared
correlation coefficient with a low root-mean-square error, and
they improve the MLR prediction model in 24%.
NTRODUCTION
A drug to be introduced into the body has to overcome
numerous barriers. For this, it undergoes through an absorption
process through several semipermeable biological membranes
to finally reach the blood. Bloodstream distributes the drug
throughout the body, where it is metabolized or eliminated by
excretion [1]. Therefore, absorption is a pharmacokinetic
process of a molecule entrance into the body to subsequently
reach systemic circulation. Biological membranes act as a filter
of different compounds which penetrate into cells. Thus,
understanding the mechanisms of drugs transport across
biological membranes, as skin, becomes necessary [2].
The main function of human skin is to protect the body
from foreign substances. There have been several studies to
increase transdermal absorption of drugs, but human skin
obtainment represents an obstacle. Synthetic polymer
membranes, such as Polydimethylsiloxane (PDMS), have been
widely employed as a substitute of human skin by many
researchers in the evaluation of transdermal drug delivery
systems [3]. Mouse skin has been widely employed in
various drug release studies. Those studies have shown that
diffusion rates of PDMS membranes are related to the
absorption through the skin [4,5]. Therefore, PDMS
membranes provide accurate and viable
MATHERIAL AND METHODS
Polar compounds pass slower than non-polar compounds
through PDMS membranes. Moreover, ionic compounds do
not pass through the membrane [5]. If diffusion is membrane
controlled and diffusant concentration in the solvent is
maintained at a negligible level, steady-state flux could be
calculated. On the other hand, steady-state flux reaches a
maximum value when the donor concentration is kept at the
solubility limit [5]. In this case, membrane permeability and
diffusant solubility values are comparable. Nevertheless,
permeability value is complicated to calculate. Furthermore, a
drug flux across the membrane is difficult to determinate and
parameter values are also complicated to obtain [5,8].
Previously mentioned inconvenient could be avoided by
using Artificial Neural Networks (ANNs) models, which are
capable of solving highly nonlinear problems. ANNs have been
extensively applied for solving several problems in various
areas [9-12], providing successfully results on complex
problems.
CONCLUSIONS
Different models have been developed with four and five
neurons in the input layer (CHA, Gb, HDCA, WPSA-2 and
Distance). All models have developed a better fit than the MLR
model developed by Weiping et al. [5] improving the model
between 13.5% and 33.02%, in Global Error terms.
The predictive power decreases, clearly with the number of
validation cases, going from 26% for models with 0%
validation cases to 14%-23% for models with 33% validation
cases. All models show a good fit in terms of RMSE and %
Error for training and validation phases. The results indicate
that neural networks have a good predictive capacity for log J
values.