INFORMATION TECHNOLOGY

Raman-algorithm linkage: cracking the code of early cancer

As a non-invasive detection technology that can obtain molecular information about diseases from body fluids, liquid biopsy for cancer has been widely used in a variety of cancers. However, the current medical clinical liquid biopsy technology relies on biospecific modification steps, resulting in high cost and time-consuming, which makes it difficult to truly apply to the universal cancer screening. In recent years, surface-enhanced Raman scattering (SERS) has emerged in cancer-related applications due to its high sensitivity technology. Among them, although the tag SERS has high accuracy, it needs to rely on biospecific modification; Although labelless SERS is inexpensive, it is difficult to avoid problems such as disturbance of the microenvironment of body fluids. Therefore, the development of a SERS technology that takes into account both high precision and low cost plays a key role in its application in clinical medicine, which is a research difficulty and hot spot at present.

Based on the above problems, Professor Xiao Xiangheng of Wuhan University, Professor Wang Fubing of Zhongnan Hospital of Wuhan University and researcher Hu Jing of University of Electronic Science and Technology of China cooperated to establish a label-free, liquid-phase optical detection system with low-cost silver nanowires as SERS probes, and carried out follow-up data analysis and cancer prediction (SERS-AICS) with the help of artificial intelligence. Through further clinical trials, we have the potential for early screening of stage I/II cancer and precancerous diseases (with an average accuracy of 88.39%) (average accuracy of up to 95.81%, average sensitivity of 95.40%, and average specificity of 95.87%) of serum samples from five different cancers (samples from 382 healthy individuals and 1582 patients).

This technology is expected to be used for low-cost, rapid and accurate universal cancer screening. More importantly, the study found that the Raman signature dimension obtained through dimensionality reduction in blood samples is closely related to cancer and reflects more microscopic changes in molecular bond energy, which makes SERS-AI a potential source of a full-spectrum omics database covering known cancer macromolecular markers and even predicting potential cancer markers in the future (Figure 1).

Figure 1: Schematic diagram of SERS-AI for high-precision serum pan-cancer screening and early screening

The article was published in the new issue of eLight of the Excellence Program under the title “Early cancer detection by serum biomolecular fingerprinting spectroscopy with machine learning.”

After obtaining the sample SERS spectral data information, the research team first selected lung cancer (LC) samples with the largest sample size among the five cancers for SERS-AI analysis, divided the training set and the test set at a ratio of 8:2, and then successfully distinguished 244 LC samples and 324 healthy control group (HC) samples from the training set and internal test set with the help of the binary classification SVM classifier, and the corresponding LC/HC model identified the area under the ROC curve of the working characteristics of lung cancer patients and healthy controls (AUC) is 0.90 (Figure 2A). Based on the good universality of SERS-AI for the recognition and classification of serum, the research team further successfully generalized it to the analysis of colorectal cancer (CRC), gastric cancer (GC), liver cancer (HCC), esophageal cancer (ESCA) and even mixed group (Mix) (Figure 2B-F). For cancer samples and normal samples, all datasets showed satisfactory differentiation, and the confusion matrix showed that the overall accuracy, sensitivity and specificity of cancer samples were as high as 95.81%, 95.87% and 95.40%, respectively (Figure 2G-I).

Figure 2: SERS-AICS-based screening for five high-mortality cancers

In addition, early cancer screening is of great significance to the survival rate of patients, so the research team is also committed to using SERS-AI technology to effectively distinguish stage I/II cancer samples from samples from other diseases in related tissues. The results showed that the common diseases related to lung, colorectal, gastric and liver/early cancer group had good model fit (Figure 3B-E), and had high accuracy, sensitivity and specificity for the identification of early cancer. Among them, this technology has the best recognition effect on gastric and liver-related datasets, corresponding to 93.33% and 92.31% accuracy, sensitivity of 100% and 85.71%, and specificity of 85.71% and 100% (Figure 3F). Therefore, the SERS-AICS system has great potential to be developed as an emerging detection method for effective screening of asymptomatic people to prevent their progression to advanced cancer.

Figure 3: Early screening of four representative cancers based on SERS-AICS

At present, the development of cancer screening is also limited, that is, the lack of databases to store, construct and track the individual characteristics of a large number of cancer patients, which limits the subsequent in-depth nature analysis, such as the search for common features of new cancer biomarkers. Based on the above tests and analyses, the SERS-AICS detection/analysis system collected and processed a total of 1964 serum samples, and its identification of 5 cancers has high accuracy, sensitivity and specificity compared to healthy controls. At the same time, this study uses the covariance matrix to assist the classification strategy of support vector machine, which shows unique advantages in analyzing the large sample serum Raman data of nearly 2000 cases. This shows that SERS-AICS can obtain more reliable spectral data from five representative cancers, which further provides an important guide for future cancer screening (Figure 4A).

More importantly, SERS-AICS can not only simplify the difficulty of model construction through covariance matrix dimensionality reduction, but also obtain cancer-related Raman signature dimensions (Figure 4B). Based on the attribution statistical analysis of the specific dimensions associated with cancer above, it is possible to calculate the shared Raman peak position between the five individual types of cancer, any mixed combination of different types of cancer, and the common characteristics of each type of cancer (Figure 4C), from which the common characteristics of each type of cancer can be further analyzed. Compared to current single-molecule biomarkers, these Raman signature dimensions cover the information of the vibrational spectra of all molecules in the sample, and the identification of various small changes due to cancer is more refined and accurate. This suggests that in future cancer marker exploration, the SERS-AI database is highly likely to be an important source of information for associative spectra and biomolecular properties.

Figure 4: Construction of cancer-related databases by SERS-AICS. (A) the covariance matrix reduces 1465 to 50 dimensions; (B) correlation heatmap when dimension selection; (C) A list of common dimensions for cancer at the molecular bond energy level, and the more stars prove that this characteristic dimension is applicable to more cancer types

In summary, SERS-AICS technology will provide an accurate, efficient and low-cost cancer detection method for high-risk groups and cancer patients through routine blood tests, which can be used as a preliminary screening method for cancer diagnosis tests such as imaging. In the future, the research team will strive to promote the application of SERS-AICS to expand to various types of early screening of cancer, and extend to the whole process of diagnosis, treatment and review of cancer patients, and hopes to eventually establish a large-scale data recording, retrieval and research system to lay a solid foundation for subsequent in-depth research and development. (Source: China Optics WeChat public account)

Related paper information:https://doi.org/10.1186/s43593-023-00051-5

Special statement: This article is reproduced only for the need to disseminate information, and does not mean to represent the views of this website or confirm the authenticity of its content; If other media, websites or individuals reprint and use from this website, they must retain the “source” indicated on this website and bear their own legal responsibilities such as copyright; If the author does not wish to be reprinted or contact the reprint fee, please contact us.



Source link

Related Articles

Leave a Reply

Your email address will not be published. Required fields are marked *

Back to top button