Fast and automated biomarker detection in breath samples with machine learning

Area of Science:

• Biomedical Engineering
• Analytical Chemistry
• Computational Biology

Background:

• Volatile organic compounds (VOCs) in human breath serve as biomarkers for various health conditions.
• Current diagnostic methods using VOCs, like gas chromatography-mass spectrometry (GC-MS), rely on time-consuming and subjective expert analysis.
• Limitations of expert-driven GC-MS analysis include potential for errors and lack of scalability.

Purpose of the Study:

• To develop and evaluate a machine learning-based system for automated GC-MS data analysis.
• To bypass the need for expert-led processing in VOC detection from breath samples.
• To improve the speed, accuracy, and consistency of breath-based diagnostics.

Main Methods:

• A machine learning system leveraging deep learning pattern recognition was proposed.
• Four types of convolutional neural networks (CNNs) were evaluated: VGG16, VGG-like, densely connected, and residual CNNs.
• The system was trained and tested on clinical breath samples to detect VOCs directly from raw GC-MS data.

Main Results:

• The machine learning approach significantly outperformed expert-led analysis in detecting VOCs.
• The AI system identified a substantially higher number of VOCs compared to traditional methods.
• The proposed methods achieved this with high specificity and in a fraction of the time.

Conclusions:

• Machine learning, particularly deep learning CNNs, offers a powerful alternative for analyzing GC-MS data.
• Automated VOC detection via AI can enhance the accuracy, speed, and consistency of breath diagnostics.
• This approach holds promise for the widespread adoption of non-invasive, cost-effective breath-based disease screening.