A supervised machine learning tool to predict the bactericidal efficiency of nanostructured surface

Abstract

Abstract The emergence and rapid spread of multidrug-resistant bacterial strains is a growing concern of public health. Inspired by the natural bactericidal surfaces of lotus leaves and shark skin, increasing attention has been focused on the use of mechano-bactericidal methods to create surfaces with antibacterial and/or bactericidal effects. There have been several studies exploring the bactericidal effect of nanostructured surfaces under various combinations of parameters. However, the correlation and synergies between these factors still need to be clarified. Recently machine learning (ML), which enables prediction or decision-making based on data, has been used in the field of biomaterials with promising results. In this study, we explored ML in nanotechnology to investigate the antimicrobial potential of nanostructured surfaces. A dataset of nanostructured surfaces and their antimicrobial properties was built by extracting the published literature. Based on the literature review and the distribution of our dataset, 70% bactericidal efficiency was selected as a practical benchmark for our classification model that balances stringent bactericidal performance with achievable targets in diverse conditions. Subsequently, we developed an ML classification model, which demonstrated an 81% accuracy in its predictive capability. A regression model was further developed to predict the value of bactericidal efficiency for nanostructured surfaces. Feature importance analysis of the ML models suggested that nanotopographical features have a greater influence on bactericidal properties than material properties, thus providing insight into the principles of the mechano-bactericidal effect of nanostructured surfaces. Overall, this ML model tool could help researchers to effectively select and design the parameters of the surface structure prior to experimentation, thereby improving the timeliness and reducing the number of experiments and the associated costs. Graphical Abstract

Keywords

• Machine learning
• Nanotopography
• Mechano-bactericidal activity
• Antimicrobial properties