High confidence Raman spectroscopy of tumor biomarker proteins through experimental and theoretical cross-validation.

Cancer represents a significant challenge to people's health and safety. Tumor biomarker detection plays a vital role in the precise diagnosis of cancer and finds widespread applications in cancer screening and pathological diagnosis. Existing methods for tumor biomarker detection have drawbacks such as susceptibility to false positives, complexity of operation, and high costs. As a molecular-level fingerprint spectrum, Raman spectroscopy holds promise as a rapid and accurate method for tumor biomarker protein detection. This paper presents a high-confidence Raman spectra collection method for tumor biomarker proteins based on experimental and theoretical cross-validation. On the one hand, through ultrafiltration purification of protein samples, high-confidence spectra for four tumor biomarker proteins of breast cancer were experimentally acquired. On the other hand, using first-principles Density Functional Theory (DFT), the Raman spectra of the proteins were calculated theoretically. Experimental and theoretical spectra were mutually validated, confirming differences in spectral peak characteristics and their assignments for the four biomarker proteins. We also demonstrate improvement in AI-based protein classification through theoretical-experimental cross-validation, with 7.62% accuracy gain. The method proposed in this paper is well-suited for integration with high-throughput spectral analysis algorithms based on artificial intelligence. It holds the potential for developing deep learning models constrained by biological knowledge in the field of cancer screening and tissue biopsy pathological diagnosis in the future.