Characterizing viral samples using machine learning for Raman and absorption spectroscopy.

Area of Science:

• Spectroscopy and Machine Learning in Pharmaceutical Analysis

Background:

• Machine learning (ML) offers robust methods for analyzing biological samples in the pharmaceutical industry.
• Predicting viral particle concentration is crucial for drug development and quality control.

Purpose of the Study:

• To predict the concentration of viral particles in biological samples using ML.
• To evaluate the efficacy of Raman and absorption spectroscopy in conjunction with ML for this prediction task.

Main Methods:

• Utilized convolutional neural networks (CNNs) and random forests (RFs) for prediction.
• Trained models using individual and combined Raman and absorption spectroscopy data.
• Developed novel networks for joint spectral data analysis.

Main Results:

• Achieved high prediction accuracy with R² values up to 95% using both RFs and CNNs.
• Demonstrated that combining Raman and absorption spectra significantly improved prediction accuracy compared to individual spectra.
• Validated the benefits of joint spectral analysis using principal component analysis.

Conclusions:

• ML models, particularly CNNs and RFs, are effective for predicting viral concentration from spectroscopic data.
• Joint analysis of Raman and absorption spectra offers superior prediction performance.
• The developed methodology shows potential for broader applications, including viral particle type characterization.