Harnessing Raman Spectroscopy and Machine Learning for Liquid Biopsy Classification: An Interview with 2025 Emerging Leader in Molecular Spectroscopy Award Winner Lingyan Shi

A joint study conducted by University of California San Diego (La Jolla, California) and the Korea Advanced Institute of Science and Technology (Daejeon, Republic of Korea) explores enhancing liquid biopsies—a noninvasive cancer classification tool—through machine learning analysis of Raman spectra to classify exosomes from different cancer cell lines: COLO205 (colon), A375 (skin), and LNCaP (prostate). By using principal component analysis (PCA) to extract chemically significant features from Raman spectra and then training a linear discriminant analysis (LDA) classifier, the study achieved 93.3% overall classification accuracy and high F1 scores (98.2% for COLO205, 91.1% for A375, and 91.0% for LNCaP). Additionally, the study compared lipid compositions of these exosomes, revealing unique lipid profiles per cancer type. These findings highlight the potential of combining Raman spectroscopy with machine learning for early cancer detection, real-time monitoring, and personalized treatment.

Spectroscopy spoke to Lingyan Shi of the University of California San Diego, corresponding author of the paper (1) that resulted from this study, about her team’s findings. Shi will receive the 2025 Emerging Leader in Molecular Spectroscopy Award, which recognizes the achievements and aspirations of a talented young molecular spectroscopist who has made strides early in their career toward the advancement of molecular spectroscopy techniques and applications. The award will be presented at SciX 2025, taking place from October 5 to 10 at the Northern Kentucky Convention Center in Covington, Kentucky.

What are some limitations of current cancer screening methods, like imaging and biopsies?
Current screening techniques—such as imaging and tissue biopsies—can be invasive, costly, and sometimes insufficient for detecting early-stage tumors or subtle molecular changes. Imaging lacks molecular specificity, while tissue biopsies are localized and may not capture tumor heterogeneity. Furthermore, repeated biopsies are not always feasible, limiting their utility for real-time monitoring.

What advantages do liquid biopsies offer over traditional tissue biopsies for cancer diagnostics?
Liquid biopsies are minimally invasive and allow for repeated sampling over time, making them well-suited for early detection, real-time monitoring, and tracking therapeutic response. They offer a more systemic view of disease by capturing tumor-derived components—such as exosomes and circulating nucleic acids—that reflect the heterogeneity of the entire tumor, rather than a single localized area.

What are the main types of biomarkers found in liquid biopsies, and why are exosomes considered promising?
Key biomarkers include circulating tumor DNA (ctDNA), RNA, proteins, and extracellular vesicles such as exosomes. Exosomes are particularly promising due to their abundance, stability, and the rich molecular cargo they carry—including lipids, proteins, and nucleic acids—that reflect their cell of origin. Their lipid-rich membranes also make them highly amenable to label-free optical detection.

What are the advantages of Raman spectroscopy in biomedical applications?
Raman spectroscopy is a label-free, non-destructive technique that provides rich molecular information based on the vibrational fingerprints of chemical bonds. It allows for precise, chemically specific detection of subtle biochemical changes in complex biological samples without the need for staining or labels.

What are some benefits of using machine learning algorithms for analyzing Raman spectra in cancer diagnostics?
Machine learning enhances the interpretability of complex spectral data by enabling robust classification, pattern recognition, and biomarker discovery. It can detect subtle spectral differences that may be imperceptible to the human eye, thereby improving diagnostic accuracy and enabling scalable, automated analysis.

Which wavenumber regions contributed most significantly to the variance in the exosome Raman spectra?
The regions around 700–900 cm⁻¹, 1000–1200 cm⁻¹, and 2800–3000 cm⁻¹ were most influential, corresponding to vibrations associated with lipids, proteins, and CH-stretching modes. These spectral regions highlight key compositional differences—particularly in lipid and protein contents—between exosomes from different cancer types.

What roles do cholesterol and phospholipids play in exosome biology and cancer progression, based on the lipid similarity analysis?
Cholesterol and phospholipids regulate membrane rigidity, exosome stability, and cargo sorting. Our lipid similarity analysis indicates that exosomal lipid composition reflects metabolic rewiring in cancer cells. Changes in these lipids may influence signaling pathways, membrane fusion, and tumor microenvironment interactions—processes that drive cancer progression and aggressiveness.

How did the lipid profiles of exosomes vary among the three cancer cell lines studied?
Each cancer cell line produced exosomes with distinct Raman spectral signatures, especially in lipid-associated regions. Interestingly, prostate (LNCap) and skin (A375) cancer exosomes exhibited near-identical lipid spectral profiles, with omeg-3 as the most abundant lipid. While in colon (COLO205) cancer exosomes, glycerophospholipids was the most abundant lipid. In addition, cholesterol consistently ranked in the top three for all cell lines. The profiles varied in that they reflected cell-type specific metabolic phenotypes.

Why is omega-3 25:5 an interesting lipid finding in this study, and what further research does it suggest?
It is interesting to find Omega-3 25:5 in cancer cells, since it is an unsaturated fatty acid with potential anti-inflammatory and anti-cancer properties. Its varying abundance across cancer-derived exosomes suggests it plays a role in tumor metabolism or immune modulation. Future studies could explore its value as a diagnostic or prognostic biomarker and its functional role in cancer biology.

What are the key challenges in translating this Raman spectroscopy-based method to clinical liquid biopsy samples?
Major challenges include standardizing sample preparation, ensuring reproducibility across instruments and platforms, and achieving sufficient sensitivity to detect low concentrations of exosomes in patient samples. Clinical translation will require validation on large patient cohorts and integration into existing clinical workflows, along with regulatory approval.

How could enhanced Raman techniques like surface-enhanced Raman spectroscopy (SERS) improve detection sensitivity in clinical applications?
SERS significantly amplifies Raman signals using nanostructured metallic surfaces, allowing for the detection of biomolecules at ultra-low concentrations, even down to the single-vesicle level. This could enable highly sensitive and rapid profiling of exosomal biomarkers for clinical applications.

How could Raman spectroscopy, combined with machine learning, impact early cancer detection and personalized treatment strategies?
Together, Raman spectroscopy and machine learning offer a powerful platform for fast, real-time, molecular-level diagnostics. They enable early detection through the identification of subtle biochemical changes and facilitate accurate cancer classification and therapeutic response prediction—key steps toward truly personalized cancer care.

What potential does this approach have in classifying cancer subtypes, such as those found in breast cancer?
Raman spectral signatures reflect biochemical differences between cancer subtypes. When paired with machine learning, this method can detect and classify these subtle variations, making it a promising tool for stratifying cancer subtypes, including breast cancer, and informing personalized treatment decisions.

How does this study contribute to the evolving landscape of noninvasive cancer diagnostics?
This study introduces a robust, label-free method for analyzing cancer-derived exosomes using Raman spectroscopy coupled with machine learning. It showcases the potential to detect cancer-specific molecular signatures from liquid biopsies, advancing the field of noninvasive early cancer detection and classification.

What are your thoughts on receiving the 2025 Emerging Leader in Molecular Spectroscopy Award?
It’s a tremendous honor to receive this prestigious award. I’m deeply grateful to my team and collaborators for their contributions. This award affirms our commitment to developing advanced spectroscopic tools for potential impactful applications in early cancer detection and precision medicine.

References

1. Villazon, J.; Dela Cruz, N.; Shi, L. Cancer Cell Line Classification Using Raman Spectroscopy of Cancer-Derived Exosomes and Machine Learning. Anal. Chem. 2025,97 (13), 7289–7298. DOI: 10.1021/acs.analchem.4c06966