Machine Learning Techniques for Effective Pathogen Detection Based on Resonant Biosensors.

Area of Science:

• Biomedical Engineering
• Machine Learning Applications
• Infectious Disease Diagnostics

Background:

• Accurate and rapid detection of SARS-CoV-2 is crucial for managing the COVID-19 pandemic.
• Optical-based biosensors offer a promising platform for virus detection.
• Signal processing is a critical step in enhancing biosensor performance.

Purpose of the Study:

• To develop and evaluate a machine learning (ML) approach for processing signals from an optical-based COVID-19 detector.
• To assess the performance of Multilayer Perceptron (MLP) and Support Vector Machine (SVM) algorithms in SARS-CoV-2 detection.
• To investigate the distinguishability of positive and negative samples using ML and data visualization techniques.

Main Methods:

• Utilized Multilayer Perceptron (MLP) and Support Vector Machine (SVM) algorithms for signal processing.
• Applied ML to both raw and feature-engineered data from the optical detector.
• Employed T-distributed stochastic neighbor embedding (t-SNE) for data visualization and analysis of sample distinguishability.
• Conducted validation with 486 negative and 108 positive samples, and control experiments with 36 negative and 732 positive samples.

Main Results:

• Achieved high performance in the qualitative detection of SARS-CoV-2.
• Demonstrated successful detection at concentrations as low as 1 TCID50/mL.
• t-SNE analysis revealed clear data distribution patterns, explaining ML prediction performance.
• ML models showed effective differentiation between positive and negative samples.

Conclusions:

• Machine learning provides a generalized and effective method for processing biosensor signals.
• The developed ML approach enhances the performance of optical-based detectors for COVID-19 diagnosis.
• This study validates the utility of ML in biosensing mechanisms reliant on resonant modes.