Semi-Selective Array for the Classification of Purines with Surface Plasmon Resonance Imaging and Deep Learning Data...

Area of Science:

• Analytical Chemistry
• Materials Science
• Biotechnology

Background:

• Real-time analysis of complex mixtures is challenging for process analytics and environmental monitoring.
• Label-free techniques like surface plasmon resonance (SPR) spectroscopy show promise but often suffer from poor analyte binding reversibility.
• Distinguishing small, chemically similar molecules in real-time requires advanced sensing strategies.

Purpose of the Study:

• To develop a robust method for identifying small, chemically similar molecules in complex samples using SPR imaging.
• To overcome the limitations of poor reversibility in traditional SPR techniques.
• To demonstrate the potential of functionalized surfaces and machine learning for enhanced molecular recognition.

Main Methods:

• Utilized SPR imaging combined with a semi-selective functional surface composed of varying ratios of graphene oxide and reduced graphene oxide.
• Developed a data analysis approach employing convolutional neural networks (CNNs) for pattern recognition.
• Investigated the sensor's ability to differentiate between four purine bases at low concentrations.

Main Results:

• The developed sensor successfully distinguished between four chemically similar purine bases.
• Classification was achieved at concentrations as low as 50 μM.
• Consistent classification accuracies were observed across multiple measurements and sensors using a standard CNN, indicating high reliability.

Conclusions:

• SPR imaging, coupled with functionalized graphene oxide surfaces and CNN-based data analysis, offers a powerful approach for identifying small, similar molecules.
• This method demonstrates high accuracy and consistency, paving the way for future online sensors in complex mixtures.
• The sensor's design addresses the challenge of poor binding reversibility, enhancing its practical applicability.