Application of artificial neural networks modeling techniques in micro/nano particle diameter estimation of TiO2 encapsulated chitosan

Microparticles and nanoparticles are used in the biomedical field for various purposes such as diagnosing, monitoring, treating and preventing diseases. Chitosan is a natural polysaccharide. In recent years, chitosan and its derivative biomaterials have attracted attention in the biomedical field thanks to their properties such as biocompatibility and low toxicity. TiO2 has various advantages in terms of its good chemical stability, non-toxicity, low cost, high photocatalytic activity and antibacterial properties. Artificial neural networks (ANN), which is a non-linear multivariate modeling, are computer programs inspired by the human brain. ANNs are structures that can provide predictions and information about samples that have never been seen by using previously learned or classified information. In this study, it was aimed to investigate the antibacterial properties of TiO2-doped chitosan micro/nano particles used in biomedical application areas and the feasibility of an artificial neural network model that can successfully predict the particle size by defining the process parameters in particle production. Particles were produced by syringe dropping method using the ionic gelation mechanism. In the presence of chitosan and chitosan/TiO2 nanoparticles in E. Coli medium, a zone formation was detected from 6 mm disc diameter (blank disc) to 7.5 mm and 8 mm disc diameter, respectively. In this case, it was determined that chitosan nanoparticles could show antimicrobial activity in the presence of TiO2. It was determined that the test output value performance graphs of the particles in all calculated groups (chitosan, chitosan/TiO2, chitosan-chitosan/TiO2) showed good fit in the MLP model of the artificial neural networks. In the MLP model, the highest R and R2 values (0.89219 and 0.8104, respectively) were obtained from chitosan/TiO2 particle data. It has been understood that the performance graphs of the expected values in all calculated groups are completely compatible with the artificial neural networks and the RBF model (R2=1). It was determined that both models were successful in size estimation in all groups, but the R and R2 values obtained from the RBF model were higher and the error values were lower.