Machine learning to predict morphology, topography and mechanical properties of sustainable gelatin-based electrospun scaffolds

Roldan Ciudad, Elisa ORCID: https://orcid.org/0000-0002-7793-7542, Reeves, Neil D ORCID: https://orcid.org/0000-0001-9213-4580, Cooper, Glen ORCID: https://orcid.org/0000-0002-9568-8973 and Andrews, Kirstie ORCID: https://orcid.org/0000-0002-0860-0604 (2024) Machine learning to predict morphology, topography and mechanical properties of sustainable gelatin-based electrospun scaffolds. Scientific Reports, 14. 21017.

Preview

Published Version Available under License Creative Commons Attribution. Download (1MB) | Preview

Abstract

Electrospinning is an outstanding manufacturing technique for producing nano-micro-scaled fibrous scaffolds comparable to biological tissues. However, the solvents used are normally hazardous for the health and the environment, which compromises the sustainability of the process and the industrial scaling. This novel study compares different machine learning models to predict how green solvents affect the morphology, topography and mechanical properties of gelatin-based scaffolds. Gelatin-based scaffolds were produced with different concentrations of distillate water (dH2O), acetic acid (HAc) and dimethyl sulfoxide (DMSO). 2214 observations, 12 machine learning approaches, including Generalised Linear Models, Generalised Additive Models, Generalised Additive Models for Location, Scale and Shape (GAMLSS), Decision Trees, Random Forest, Support Vector Machine and Artificial Neural Network, and a total of 72 models were developed to predict diameter of the fibres, inter-fibre separation, roughness, ultimate tensile strength, Young’s modulus and strain at break. The best GAMLSS models improved the performance of R2 with respect to the popular regression models by 6.868%, and the MAPE was improved by 21.16%. HAc highly influenced the morphology and topography; however, the importance of DMSO was higher in the mechanical properties. The addition of the morphological properties as covariates in the topographic and mechanical models enhanced their understanding.