Lignocellulosic materials, the essential components of plant matter, are organized in complex three-dimensional structures, where detailed understanding of the interactions between the different polymer components is still an unresolved challenge. In this work, a novel microscopy imaging and artificial intelligence infrastructure is used to visualize individual lignocellulosic building blocks and their assembled structures at an unprecedented resolution. The application potential is in gaining information to improve biorefinery processes - serving in the development of selective dissolution processes for producing purified streams of cellulose, lignin and hemicelluloses. It can also be applied in new technologies for producing natural textile fibers and to reveal the detailed surface structure of lignocellulosic materials, essential for enzyme-aided or chemical modification in biorefineries.
We have demonstrated visualization of individual lignocellulosic molecules and their assembled structures at an unprecedented resolution.
We have demonstrated that our machine learning method based on a Convolutional Neural Network architecture can identify adsorption configurations accurately. On a complex system, such as bio-based molecules, this allows us to drastically reduce the number of possible molecular solutions from a set of experimental images – rapid characterization of atomic positions, elemental character and electrostatic properties is now possible.
We have established a protocol for the isolation, preparation, characterization and microscopy measurements of new bio-based samples integrated to a workflow for machine learning predictions.
Oinonen, N., Xu, C., Alldritt, B., Federici Canova, F., Urtev, F., Cai, S., Krejčí, O., Kannala, J., Liljeroth, P., and Foster, A.S. Electrostatic Discovery Atomic Force Microscopy. ACS Nano, 16 (1), 89-97 (2022). https://doi.org/10.1021/acsnano.1c06840
Oinonen, N., Kurki, L., Ilin, A., and Foster, A.S. Molecule graph reconstruction from atomic force microscope images with machine learning. MRS Bulletin, 47, 895–905 (2022). https://doi.org/10.1557/s43577-022-00324-3
Yurtsever, A., Wang, P.-X., Priante, F., Morais Jaques, Y., Miyata, K., MacLachlan, M.J., Foster, A.S., and Fukuma, T. Probing the Structural Details of Chitin Nanocrystal–Water Interfaces by Three-Dimensional Atomic Force Microscopy. Small Methods, 6(9), 2200320 (2022). https://doi.org/10.1002/smtd.202200320
Ranawat, Y.S., Jaques, Y.M., and Foster, A.S. Generalised deep-learning workflow for the prediction of hydration layers over surfaces. Journal of Molecular Liquids, 367, part B, 120571 (2022). https://doi.org/10.1016/j.molliq.2022.120571