Can Machine Learning Reduce AFM Uncertainty?
Two studies published in 2022 demonstrate machine learning techniques used to reduce uncertainty in atomic force microscopy (AFM). Making AFM more accurate through artificial intelligence (AI) could make the technology more accessible worldwide, as the algorithmic advantage these approaches provide can yield reliable results from surveys with less accurate instrumentation. mechanically.
Image Credit: TierneyMJ/Shutterstock.com
AFM devices detect the atomic forces between an extremely sharp pointed probe and the sample being studied. The probe is attached to a cantilever and at the other end is a very sensitive sensor.
The sensor records nanoscale deviations in the cantilever which provide information about the surface nanostructure of the sample.
AI refers to computer processes that perform a task in a way that appears intelligent. They accomplish this with large amounts of data and processing power and algorithms that tell the computer what to do with the data at hand.
Machine learning is a classic example of AI. Machine learning algorithms can define their own subtasks and additional parameters to better perform the tasks originally defined for them.
Machine learning can be an appropriate way to combat uncertainty in AFM. Uncertainty necessarily enters into AFM calculations and simply refers to the margin of error for which the instrumentation system is rated.
Manufacturers and researchers report uncertainty as the mean and standard deviation of a number of repeated experiments.
Machine learning methods are well suited to the task of reducing uncertainty due to their ability to improve iteratively with multiple replicated data points to work with over time.
Two recent studies have explored machine learning approaches to reduce uncertainty in AFM.
Reduced uncertainty of AFM nanomechanical measurements with appropriate contact model selection
In an article published in the European Journal of Mechanics In 2022, nanofabrication and nanoanalytical engineers from National Cheng Kung University, Tainan, Taiwan, proposed a supervised machine learning (SML) algorithmic framework to help reduce uncertainty in AFM measurements.
The machine learning method worked by selecting an appropriate contact pattern; in other words, establish the best parameters to evaluate the accuracy of AFM measurements.
The SML technique used the selected contact model to evaluate the modulus of elasticity based on the characteristics of the force curve recorded in the AFM instrument. They extracted these features for the SML classification algorithm. This algorithm then learns the appropriate contact pattern by analyzing these outputs.
Five distinct classifiers were used alongside four contact models. These were trained on (i.e. the algorithms retrieved information from and learned from) samples of polyvinyl alcohol, polymethyl methacrylate, polydimethylsiloxane, and gold.
One of the classifiers, for Linear Discriminant Analysis (LDA), emerged as the top performer in terms of providing good quality predictions with its contact models for the tested materials.
The team treated Staphylococcus aureus bacteria to obtain novel data to demonstrate the practical viability of their structure. With the LDA classifier, they achieved a test accuracy of 96.8%.
The researchers concluded that the machine learning classifier they developed was a potentially powerful tool for selecting appropriate contact patterns to reduce uncertainty in AFM investigations, even when no prior knowledge is available. .
Predict Electrochemical Impedance Spectra with AI-Enabled AFM
Another article published in 2022, in the journal Scientific reportsdemonstrated a convolutional neural network (CNN) developed to reduce uncertainty in AFM investigations of supported bilayers.
AFM analysis of tethered bilayer membranes (tBLM), for example, helps scientists understand how pore-forming proteins damage membranes at the nanoscale. This data can then be used to predict electrochemical impedance spectroscopy (EIS) of these objects.
EIS is a complicated technique used to analyze bio-recognition events that occur on the surface of electrodes. EIS analyzes interfacial properties related to events such as antibody-antigen recognition, whole-cell capture, and enzymatic substrate interaction. It has potentially impactful applications in environmental science as well as biomedicine.
The recent paper, written by an interdisciplinary team of computer scientists and biochemists working at Vilnius University, Lithuania, presented a CNN that predicted the EIS response of tBLMs with a finite element analysis modeling technique ( AEF).
CNN is a deep learning technique and a type of artificial neural network (ANN). FEA is a method for solving differential equations.
The team found that their method predicted electrochemical impedance spectra well enough.
Nanoscale Data, Automation and Instrumentation
Advances towards smaller, more efficient and more powerful electronic devices – with nanoelectronics, spintronics and nanoscale fabrication techniques beginning to mature – will continue to generate more data from more accurate and applicable sensors. on nanoscale instruments such as AFM devices.
When this is combined with another technology trend – towards more automation – the increase in data sources is compounded. More automation leads to more actions taken, more records scored and shared, and more sets analyzed for trends.
Ultimately, this will continue to reduce measurement uncertainty for nanoscale investigative instruments.
References and further reading
Magar, HS, RYA Hassan and A. Mulchandi (2021). Electrochemical impedance spectroscopy (EIS): principles, construction and applications of biosensing. Sensors. doi.org/10.3390/s21196578.
Nguyen, LTP and BH Liu (2022). Machine learning approach to reduce uncertainty in AFM nanomechanical measurements through the selection of the appropriate contact model. European Journal of Mechanics – A/Solids. https://www.sciencedirect.com/science/article/pii/S0997753822000626?via%3Dihub
Raila, T. et al. (2022). AI-based atomic force microscopy image analysis predicts electrochemical impedance spectra of defects in tethered bilayer membranes. Scientific reports. https://www.nature.com/articles/s41598-022-04853-4.