Machine learning for detecting DNA attachment on SPR biosensor.

Area of Science:

• Biomedical Engineering
• Biosensor Technology
• Machine Learning Applications

Background:

• Optoelectric biosensors, particularly surface plasmon resonance (SPR)-based sensors, are crucial for label-free detection of biomolecular interactions in biomedical diagnostics.
• While SPR biosensors offer high precision, there's a need for robust machine learning (ML) models to assess their accuracy and ensure reliable datasets for disease diagnosis.
• Current ML models often lack specific validation for SPR biosensor data, hindering downstream applications.

Purpose of the Study:

• To develop and evaluate innovative machine learning models for DNA detection and classification using data from SPR biosensors.
• To assess the performance of various ML classifiers on SPR-generated datasets, focusing on reflective light angles and gold surface properties.
• To establish reliable methods for evaluating SPR biosensor accuracy and ensuring data quality for future diagnostic tool development.

Main Methods:

• Utilized reflective light angles and gold surface properties from SPR biosensors as input features for ML models.
• Applied t-distributed Stochastic Neighbor Embedding (t-SNE) for feature extraction and min-max normalization to differentiate low-variance classifiers.
• Experimented with multiple ML classifiers including Support Vector Machine (SVM), Decision Tree (DT), Multi-layer Perceptron (MLP), K-Nearest Neighbors (KNN), Logistic Regression (LR), and Random Forest (RF).

Main Results:

• Achieved high accuracy for DNA classification (0.94) using Random Forest (RF), Decision Tree (DT), and K-Nearest Neighbors (KNN).
• Reached superior accuracy for DNA detection (0.96) with Random Forest (RF) and K-Nearest Neighbors (KNN).
• Random Forest (RF) demonstrated the best overall performance with an Area Under the Curve (AUC) of 0.97, precision of 0.96, and F1-score of 0.97 for both detection and classification tasks.

Conclusions:

• Machine learning models show significant potential in advancing SPR biosensor development and performance evaluation.
• The developed ML models provide a reliable framework for assessing SPR biosensor accuracy and data quality.
• This research paves the way for creating novel diagnostic and prognostic tools for various diseases based on enhanced biosensor technology.