Detecting Respiratory Viruses Using a Portable NIR Spectrometer-A Preliminary Exploration with a Data Driven...

Area of Science:

• Analytical Chemistry
• Biotechnology
• Machine Learning Applications

Background:

• Accurate detection of respiratory viruses is crucial for pandemic management.
• Conventional laboratory methods are costly and time-consuming.
• Portable Near-Infrared (NIR) spectroscopy offers a low-cost, rapid, and field-deployable alternative, but faces challenges with specificity and data quality.

Purpose of the Study:

• To develop and validate a machine learning-enhanced portable NIR spectroscopy method for detecting respiratory syncytial virus (RSV) and Sendai virus (SEV).
• To overcome the limitations of low specificity and interweaving spectral features in NIR spectroscopy through advanced data analysis.

Main Methods:

• Utilized a portable NIR spectrometer for sample analysis.
• Implemented a machine learning approach incorporating variable selection via Variable Importance in Projection (VIP) scores and quantile values.
• Employed variable truncation processing to enhance model accuracy.
• Conducted extensive experiments using four datasets with varying training, validation, and testing splits.

Main Results:

• Achieved high classification accuracy for RSV, SEV, and combined detection across different experimental setups.
• Average accuracies reached up to 0.94 for RSV, 0.97 for SEV, and 0.97 for RSV + SEV during model validation.
• Model testing yielded average accuracies of 0.90 (RSV), 0.93 (SEV), and 0.91 (RSV + SEV).

Conclusions:

• Portable NIR spectroscopy, when enhanced with a sophisticated machine learning algorithm, demonstrates significant feasibility for accurate respiratory virus detection.
• The developed approach offers a viable solution for rapid population screening and early detection, contributing to pandemic preparedness.
• This method addresses the limitations of traditional techniques by providing a cost-effective and highly deployable diagnostic tool.