Scholarly record

APPLICATION OF NEURAL NETWORK MODEL FOR PREDICTING THE ANTIBACTERIAL ACTIVITY OF ALGINATE-CHITOSAN SPONGES

М. Г. Гордиенко

Abstract

Medical materials based on natural polysaccharides can be used as hemostatic sorbents and dressings. In this work, some samples of sponges had been obtained based on sodium alginate and chitosan. The concentration of sodium alginate in the solution, the ratio of alginate and chitosan were varied during an experiment. Polysaccharide cross-linking was carried out by inotropic gelation or by combining ionotropic gelation with ionic substitution. According to the literature, a chitosan have antibacterial properties against many microorganisms. To enhance the antibacterial effect of sponges, a silver nanoparticles were incorporated into some samples. The silver nanoparticles had been obtained by microbiological synthesis by using three different cultures of fungi: F.nivale, F.oxysporum and P.glabrum. The antibacterial activity of sponges was tested on three types of microorganisms: B.cereus, S.aureus and P.aeruginosa. The results did not allow us to reveal a direct correlation between the composition of the sponge and the width of the zone of inhibition. To predict antibacterial activity, the use of a neural network model has been proposed. To build a model, an array of data containing information on 48 samples was divided into test and training datasets. Were built 3 neural networks (separately for each type of microorganisms) with a different number of hidden layers and the number of neurons in them. Network training was performed by back propagating the error. Validation of the models was performed on a test sample. The standard deviation was chosen as the criterion. In general, satisfactory results were obtained only for the culture of S.saureus. For this culture, with the best result, the network 4-4-1 gave an error on the training and test sample of 10%; and the network 4-2-3-1 ? 15%. To increase the accuracy of the model is required to increase the amount of data.

Publication Impact Profile

• Captures
• Mendeley - Readers: 6

PLUMX see details

DOI resolver: doi.org/10.5593/sgem2018v/6.4/s08.010

Publication details

Back to publication list