Related Experiment Video
Updated: Oct 13, 2025
09:27
Automation of Bio-Atomic Force Microscope Measurements on Hundreds of C. albicans Cells
Published on: April 2, 2021
4.0K
Supervised Learning for Predictive Pore Size Classification of Regenerated Cellulose Membranes Based on Atomic Force
Alex Hadsell1, Huong Chau1,2, Richard Barber2,3
1Department of Bioengineering, Santa Clara University, Santa Clara, CA 95053, USA.
Materials (Basel, Switzerland)
|November 13, 2021
Summary
This study developed a supervised learning algorithm to classify nanoporous dialysis membrane pore sizes. The algorithm accurately distinguishes larger pores (1000 kDa vs. 100 kDa) but struggles with smaller ones (100 kDa vs. 50 kDa) due to instrument limitations.
Area of Science:
- Biomaterials Science
- Nanotechnology
- Analytical Chemistry
Background:
- Regenerated cellulose nanoporous membranes serve as molecular weight cutoff standards in bioseparations.
- Accurate characterization of these membranes is crucial for developing advanced biomimetic separation technologies.
Purpose of the Study:
- To develop and validate a supervised learning algorithm for classifying nanoporous membrane pore sizes.
- To assess the performance of atomic force microscopy (AFM) in pore size characterization for machine learning applications.
- To optimize classification accuracy and reduce scan times for membrane analysis.
Main Methods:
- Mesoporous standards (50, 100, 1000 kDa) characterized using Atomic Force Microscopy (AFM).
- Supervised learning approach employing Gamma transformation, discriminant analysis (Area Under the Curve - AUC, Accuracy - Acc), and logistic regression.
- Monte Carlo simulations for dataset generation and WEKA software for algorithm validation.
Main Results:
- The algorithm achieved high accuracy (AUC > 0.8) in classifying 1000 kDa versus 100 kDa membranes.
- Discrimination between 100 kDa and 50 kDa membranes was weaker (AUC < 0.7), attributed to AFM instrument accuracy limitations below 5 nm.
- Cross-validation indicated that 70-80% of training data is required for accurate classification, suggesting potential for scan time reduction.
Conclusions:
- Supervised learning, combined with AFM, offers a viable method for classifying nanoporous membrane pore sizes.
- Instrument accuracy is a critical factor limiting the resolution of pore size classification for smaller membranes.
- The developed approach shows promise for optimizing membrane characterization processes and accelerating the development of biomimetic membranes.
