Banca de DEFESA: MARIANA RODRIGUES VILLARIM

Uma banca de DEFESA de DOUTORADO foi cadastrada pelo programa.
STUDENT : MARIANA RODRIGUES VILLARIM
DATE: 26/06/2025
TIME: 09:00
LOCAL: Remoto (Google Meet: http://meet.google.com/kis-jpuz-xhf)
TITLE:

DEVELOPMENT OF A COMPUTATIONAL TOOL FOR SURFACE PLASMON RESONANCE-BASED BIOSENSORS USING HYBRID METHODS AND MACHINE LEARNING

KEY WORDS:

Surface Plasmon Resonance, Biosensors, Mathematical Modeling, Hybrid Spectro-Angular Analysis, Curve Smoothing Techniques, Machine Learning, Computational Analysis.

PAGES: 82
BIG AREA: Engenharias
AREA: Engenharia Elétrica
SUBÁREA: Materiais Elétricos
SPECIALTY: Materiais e Componentes Semicondutores
SUMMARY:

The optical biosensor based on Surface Plasmon Resonance (SPR) offers high sensitivity, label-free operation, and its multilayer construction enables greater selectivity to the target analyte. These devices are widely used in fields such as healthcare, environmental monitoring, the food industry, and agriculture. Initially, a mathematical and numerical model was developed with three layers (prism, metal, and sensing medium) and compared using the transfer matrix and finite difference methods.
A computational tool was developed using MATLAB App Designer, enabling simulations, experimental data import, real-time graphical visualization, and result export. The application allows users to analyze different configurations of the SPR biosensor through an intuitive and accessible interface. Among the main technical advancements implemented are: (i) integration of traditional and hybrid analysis modes (angle versus wavelength); (ii) application of advanced curve smoothing techniques; and (iii) implementation of machine learning algorithms to predict the minimum resonance angle even in situations with incomplete data. The multidimensional hybrid analysis contributes to a more robust optimization of biosensor parameters, with significant reduction in spectral noise and greater consistency in results under varying experimental conditions. Furthermore, by enhancing the multilayer structure with different chemical components, an increase in sensitivity of 12.99% for silver and 16.59% for gold was observed.
A comparison of the curve smoothing techniques applied showed that the Savitzky-Golay filter and spline smoothing achieved the best performance in minimizing unwanted noise. Among the machine learning methods, Gaussian Process Regression (GPR) and Neural Networks demonstrated the best performance when applied to incomplete data—relevant in experimental applications where data acquisition may be limited by noise or failures—achieving very high correlation values, lower standard deviations, and reduced absolute and mean squared errors.
Therefore, the developed tool and the results obtained demonstrate the potential of the system as a viable alternative for the analysis and optimization of SPR biosensors, contributing directly to their scientific and technological advancement.
Finally, the assembly of a compact and portable physical structure was initiated, aiming at the generation of a database.

COMMITTEE MEMBERS:
Presidente - 2140683 - DIOMADSON RODRIGUES BELFORT
Interno - 1284113 - SEBASTIAN YURI CAVALCANTI CATUNDA
Externo à Instituição - JOYLAN NUNES MACIEL - UNILA
Externo à Instituição - MARCO ROBERTO CAVALLARI - UNICAMP
Externo à Instituição - JOÃO PAULO PEREIRA DO CARMO - USP
Externo à Instituição - OSWALDO HIDEO ANDO JUNIOR - UFRPE