Deep learning aided quantitative analysis of anti-tuberculosis fixed-dose combinatorial formulation by terahertz...

Area of Science:

• Analytical Chemistry
• Spectroscopy
• Machine Learning

Background:

• Accurate quantification of active ingredients in anti-tuberculosis fixed-dose combinations (FDCs) is crucial for treatment efficacy.
• Insufficient drug content can lead to reduced curative effects and the development of drug-resistant tuberculosis strains.
• Quality control of FDCs relies heavily on precise detection of active component concentrations.

Purpose of the Study:

• To develop a novel quantitative calibration model for detecting active component content in anti-tuberculosis FDCs.
• To evaluate the performance of Convolutional Neural Networks (CNN) for this application using terahertz time-domain spectroscopy (THz-TDS).
• To compare the efficacy of CNN with traditional Partial Least Squares Regression (PLSR) methods.

Main Methods:

• Utilized Convolutional Neural Networks (CNN), a deep learning technique, for quantitative analysis.
• Employed terahertz time-domain spectroscopy (THz-TDS) to acquire spectral data.
• Developed a reference model using Partial Least Squares Regression (PLSR) with various data preprocessing techniques for comparison.

Main Results:

• CNN models directly processed raw terahertz spectral data.
• PLSR models required extensive data preprocessing to improve prediction performance.
• CNN models demonstrated superior prediction accuracy compared to PLSR, even with preprocessed data.

Conclusions:

• Convolutional Neural Networks (CNN) offer an ideal and highly effective method for the quantitative analysis of active components in anti-tuberculosis FDCs.
• The direct application of CNN to raw spectral data simplifies the analysis process.
• This approach enhances the quality control of tuberculosis medications and aids in combating drug resistance.