Speaker
Description
Atomic force microscopy force spectroscopy (AFM-FS), in which the AFM tip is pressed into and retracted from a surface to generate force-distance curves, can be used to obtain high resolution maps of the topography and mechanical properties of materials. This technique generates a large amount of data where each pixel in an image corresponds to a single force-distance curve. AFM-FS has been widely used to characterize lateral inhomogeneities in samples such as composite materials, polymer blends, and particles on a surface. With inhomogeneities on small, nanoscale length scales, it is challenging to resolve and quantify the properties of the different regions. By engineering several features from force-distance curves that differentiate the mechanical properties of the different regions, including stiffness, adhesion, and indentation depth, we have trained machine learning classifiers to identify phytoglycogen (PG) nanoparticles on a surface as well as regions of different mechanical properties within individual PG nanoparticles [1]. PG is produced in the kernels of sweet corn as soft glucose-based nanoparticles with an underlying dendritic architecture. AFM-FS measurements of these samples in water result in force-distance curves with different features due to the difference in the mechanical properties of the soft nanoparticles and the surrounding hard substrate. In addition, because of their dendritic architecture, there are variations in the mechanical properties within individual PG nanoparticles, which can be interpreted as a softer outer region and a stiffer inner region [2]. By training supervised and unsupervised machine learning classifiers to identify the soft nanoparticles on the hard surface as well as to distinguish the stiffer inner region and softer outer region within each nanoparticle, we have quantified the nanoscale properties of the nanoparticles, such as the particle modulus, the difference in modulus between the inner and outer regions, and the length of the exterior glucose chains. Our approach highlights the advantages of machine learning classifiers as a powerful tool in identifying and quantifying different regions within laterally inhomogeneous samples, and it shows promise for application to other soft matter and biological systems.
[1] B. Baylis et al., Soft Matter (2026).
[2] B. Baylis et al., Biomacromolecules 22, 2985 (2021).
| Keyword-1 | Machine Learning |
|---|---|
| Keyword-2 | Atomic Force Microscopy |
| Keyword-3 | Soft Matter |