End stage renal disease (ESRD), characterized by cessation in kidney function, has been linked to severe metabolic disturbances, caused by buildup of toxic solutes in blood. To remove these solutes, ESRD patients undergo dialysis. As a proof of concept, we tested whether ESRD-related metabolic signatures can be detected in perspiration samples using a combined methodology. Our rapid methodology involves swabbing a glass slide across the patient’s forehead, detecting the metabolites in the imprint using desorption electrospray ionization mass spectrometry, and identifying the key differences using machine learning methods. Based on collecting 42 healthy and 27 ESRD samples, we find saturated fatty acids are consistently suppressed in ESRD patients, with little change after dialysis. Also, our method enables the detection of uremic solutes, where we find elevated levels of uric acid (6.7 fold higher on average) that sharply decrease after dialysis. Beyond the study of individual metabolites, we find that a lasso model, which selects for 8 m/z fragments from 24,602 detected analytes, achieves area under the curve performance of 0.85 and 0.87 on training (n=52) and validation sets (n=17) respectively. Together, these results suggest that this methodology is promising for detecting signatures relevant for Precision Health.

Background: Oral squamous cell carcinoma (OSCC) accounts for 90 % of oral cancers. If a necessary intervention before tumorigenesis could be conducted, the current 60% 5-year survival rate would be expected to be majorly improved. This fact motivates the search for developing a highly sensitive and specific in vitro diagnostic method to conduct rapid OSCC screening. Method: Serum samples from 819 volunteers, consisted of 241 healthy contrast (HC) and 578 OSCC patients, were collected, and their metabolic profiles were acquired using conductive polymer spray ionization mass spectrometry (CPSI-MS). Univariate analysis was used to select significantly changed metabolite ions in the OSCC group compared to the HC group. Identities of these metabolite ions were determined by MS/MS experiments and reconfirmed at the tissue level by desorption electrospray ionization mass spectrometry (DESI-MS). The supporting vector machine (SVM) algorithm was employed as the machine learning model to implement the automatic prediction of OSCC. Results: Through statistical analysis, 65 metabolites were selected as potential characteristic marker candidates for serum OSCC screening. In situ validation by DESI-MSI revealed that 8 out of top 10 metabolites showed the same trends of change in tissue and serum. With the aid of machine learning, OSCC can be distinguished from HC with an accuracy of 98.0 % by cross-validation in the discovery cohort and 89.2% accuracy in the validation cohort. Furthermore, orthogonal partial least square-discriminant analysis (OPLS-DA) also showed the potential for recognizing OSCC stages. Conclusion: Using CPSI-MS combined with SVM, it is possible to distinguish OSCC from HC in a few minutes with high specificity and sensitivity, making this rapid diagnostic procedure a promising approach for high-risk population screening.