Deep Learning Applied to Raman Spectroscopy for the Detection of Microsatellite Instability/MMR Deficient Colorectal...

Area of Science:

• Biomedical Optics
• Molecular Spectroscopy
• Computational Biology

Background:

• Defective DNA mismatch repair is a key pathway in colorectal cancer development.
• Microsatellite instability (MSI) is a crucial biomarker for detecting this defect.
• Current universal testing for MSI faces resource limitations, necessitating novel detection methods.

Purpose of the Study:

• To investigate Raman spectroscopy (RS) for distinguishing between normal, microsatellite stable (MSS), and microsatellite unstable (MSI-H) colorectal adenocarcinoma.
• To evaluate the efficacy of deep learning models compared to traditional machine learning for MSI detection using RS data.
• To identify biochemical differences between sample types using RS.

Main Methods:

• Development of a 1D convolutional neural network (CNN) for classifying colorectal tissue samples.
• Comparison of the CNN with principal component analysis-linear discriminant analysis (PCA-LDA) and support vector machine (SVM) models.
• Utilized a nested cross-validation strategy with 30 samples (10 per group) and 1490 Raman spectra.

Main Results:

• The CNN achieved 83% sensitivity and 45% specificity.
• PCA-LDA demonstrated 82% sensitivity and 51% specificity.
• Raman peaks associated with nucleic acids and collagen were implicated in sample discrimination.

Conclusions:

• Raman spectroscopy, particularly when analyzed with deep learning, shows potential for accurate colorectal cancer biomarker detection.
• The findings suggest RS can effectively discriminate between MSS and MSI-H tissues, offering a viable alternative for MSI testing.
• The study highlights the molecular discriminative power of RS and implicates specific biochemical components in the detection process.