Document Type : Original Research Article
Authors
1 North Khorasan University of Medical Sciences
2 University of Tabriz
3 Tehran University of Medical Sciences
Abstract
Keywords
Main Subjects
INTRODUCTION
Electrohydrodynamic atomization (EHDA) or electrospray is being used to produce polymeric nanoparticles (NPs). In this method, liquid droplets are produced under the influence of a high electric field. The technique involves electromechanical and hydrodynamic forces, that control size and morphology of the NPs. Based on Fig. 1, electrospray apparatus consists of four parts: mechanical syringe pump, high voltage electric power supply, metal needle (nozzle) and grounded metal collector. During the electrospray process, the polymer solution is infused from the needle that is connected to the syringe pump and wired to the high voltage power supply.
Parameters that may affect size and size distribution of NPs in this method are solution parameters (e.g. type of polymer, concentration of polymer, molecular weight and type of solvent used), processing parameters (e.g. flow rate and applied voltage), as well as environmental parameters (e.g. humidity and temperature) (1, 2).
It is now well-known that size of NPs is an important parameter in determining their physicochemical properties. Many different biological properties have been reported to highly depend on size of particles. They include crossing through various biological barriers (3-5), rate of endocytosis (6) and lysosomal escape (7), response of immune system (8) as well as biodistribution in body and release profile (9, 10). Many factors have been reported to affect size of NPs, including concentration of ingredients (e.g. PLA-PEG-PLA copolymer (11), chitosan (CS) (12), CS and albumin (13), TPP (14), salt (15), surfactant (16)), polymer chain length (17), polymer molecular weight (18, 19), applied energy (20) and sonication time/ amplitude (21).
Artificial neural networks (ANNs) are computational models vaguely inspired by biological neural networks which plan to model patterns and learning capacity of the human brain (22). ANNs are usually used to find the patterns or relationships between inputs and output(s). ANNs are proper alternatives for situations where standard statistical analyses are not able to analyze complex, multi-dimensional and nonlinear patterns, or where the data are poorly organized (23). ANNs offer better validity and predictability compared with experimental designs (24). Example of their uses include toxicity prediction (25, 26), pharmacokinetics and pharmacodynamics studies (27, 28).
Reviewing the literature, some reports may be found on the effect of different independent parameters on properties of electrosprayed particles, especially their particle size. Such information would provide insight into how to control the electrospray parameters to obtain the desired particle. Effect of polymer concentration on electrosprayed nanoparticles has been reported at concentration range of 0.1-0.7 (%w/v) (29), while in this study, a wider range of concentration was investigated on CS NPs. Additionally, little to no research was found on the effect of temperature/ humidity on the size of electrosprayed CS NPs. This investigation focused on identifying the influence of four independent parameters, namely, applied electric voltage, concentration of polymer used (i.e. CS), temperature and humidity, on particle size of CS NPs prepared by electrospray.
MATERIALS AND METHODS
Materials
CS (MW=7 kDa, DD~72%) was purchased from Zhengzhou Sigma Chemical Co. (China). Tripolyphosphate (TPP) was obtained from Sigma-Aldrich (USA).
In this study, modeling the parameters affecting particle size was performed using INForm v4.02 (Intelligensys, UK). The input variables were concentration of polymer, humidity, temperature and applied voltage and the output was size of NPs.
Forty-seven CS NPs samples were experimentally prepared having different values for the four factors, while the remaining factors were fixed as mentioned in section 2.2. Their particle size was then measured using DLS. Afterwards, the data sets were randomly divided into three sets (categories) to perform the ANNs modeling. The sets included test, validation and training data. Thirty-six data were used to train network, using learning parameters/ algorithms listed in Table 1. The network structure included one hidden layer with 5 nodes. Remaining parameters were given previously (31). The software selected 10% percent of the training data as test data (four points) to check fit of the model and avoid overtraining of the chosen model (32). The remaining data sets (7 samples (were excluded from training the model and taken as validation (unseen) data (Table 2) to validate the generated model. After developing the model and training the network, predictive ability and quality of the model were evaluated using the determination coefficient (R2) for the data sets.
R2=1-
Where n was number of samples, is the mean obtained values, and ŷ represented the predicted values by the model for the dependent variable. High R2 value (close to 1) is preferred for the unseen, test and training data sets.
Afterwards, response surfaces produced by the software were used to investigate the effects and relationships between the input data and the output parameter.
CONFLICTS OF INTEREST
The authors declare that there are no conflicts of interest.