Combining Green Analytical Methods and Machine Learning to Develop Urinary Fatty Acid Models for Cancer Detection
Abstract
This study aimed to explore fatty acids (FAs) as non-invasive biomarkers for prostate cancer (PCa) detection and prognosis, and their potential applications in other cancers. The objectives of this study were: 1) Develop a non-invasive urinary FAs-based model for diagnosing PCa; 2) Develop a FAs-based liquid biopsy model for non-invasive monitoring of PCa progression; 3) Investigate FAs composition in periprostatic adipose tissue (PPAT) collected from PCa positive patients; 4) Investigate the potential of urinary FAs biomarkers for diagnosing clear cell renal cell carcinoma (ccRCC) and ovarian cancer (OC).For the PCa diagnostic model, urine samples from 334 biopsy-confirmed PCa-positive and 232 PCa-negative subjects were analyzed for FAs content using stir bar sorptive extraction coupled with gas chromatography/mass spectrometry (SBSE-GC/MS). The data was split into training (70%) and testing (30%) sets to develop and validate the logit models. The FAs-based model achieved an area under the curve (AUC) of 0.71 (95% CI = 0.67-0.75, sensitivity= 0.48, and specificity= 0.83). In comparison, the PSA model performed with an AUC of 0.51 (95% CI = 0.46-0.66, sensitivity= 0.44, and specificity= 0.71). The finding clearly shows that our FAs model for PCa diagnosis outperformed the current standard test (PSA).We further examined the use of FAs for differentiating patients of clinically significant (i.e. high risk) PCa from those with indolent PCa (i.e. low risk) for a non-invasive FAs-based liquid biopsy model for PCa prognosis. Urine samples collected from 390 biopsy-designated PCa positive patients were categorized based on their Grade Group (GG), 213 were classified as GG 1 (low-risk PCa) and 177 were classified as GG 2 to 5 (high-risk PCa). The FA model for PCa prognosis selected 32 compounds, 31 FAs and 1 sterol, and achieved an AUC of 0.76 (95% CI = 0.67-0.84), sensitivity of 0.87, and specificity of 0.48.As urinary FAs are shown to be promising biomarkers for PCa diagnosis and prognosis, the source of those FAs is of particular interest. We then turned to investigate the FAs composition in periprostatic adipose tissue (PPAT) collected from PCa positive patients. We analyzed 33 PPAT samples for their FAs content with SBSE-GC/MS. The FAs profile of PPAT consisted of a total of 279 FAs. We observed that among the PPAT samples the carbon-18 (C18) was the most abundant length chain. We also observed that 14 out of the 20 FAs in the urine PCa diagnostic model could be traced back to PPAT, and that 9 out of the FAs 32 in the urine PCa prognosis model could be traced back to PPAT. The findings suggest that many urinary FA biomarkers are originated from PPAT which was found to be related to PCa progression. The finding may validate the use of urinary FAs for cancer detection as a non-invasive alternative.Finally, we tested the application of urinary FAs biomarkers for detections of other cancers, such as clear cell renal cell carcinoma (ccRCC) and ovarian cancer (OC). We collected urine samples from 233 Computed Tomography (CT) designated ccRCC positive patients and 43 control patients. Following the same procedure for PCa models, the ccRCC diagnostic model was developed with 14 FAs and an AUC of 0.92 (95% CI = 0.81-1.00), sensitivity of 0.88, and specificity of 0.87. A separate urine cohort of 31 urine samples from 16 OC positive, 5 benign, and 10 patients was used for exploration of the application of urinary FAs in OC detection. The 31 samples of OC underwent a partial least squares-discriminant analysis (PLS-DA). The PLS-DA clearly showed distinguished clusters among the 3 OC sample types. And among the most significant variables, several FAs were found to be important contributors. The FAs were cis-7-Hexadecenoic acid (C16:1), and methyl stearate (C18:0) identified for OC classification.The study demonstrates that urinary FAs can serve as non-invasive biomarkers for PCa diagnosis and prognosis, as well as for other cancer types like ccRCC and OC. These findings highlight the potential of FA-based models to improve cancer detection and reduce the need for invasive procedures, ultimately enhancing patient care and reducing healthcare costs. Further research is needed to refine these models and expand their application in clinical practice.
Subject Area
Chemistry|Cellular biology|Oncology|Biochemistry
Recommended Citation
Noriega Landa, Elizabeth, "Combining Green Analytical Methods and Machine Learning to Develop Urinary Fatty Acid Models for Cancer Detection" (2024). ETD Collection for University of Texas, El Paso. AAI31298176.
https://scholarworks.utep.edu/dissertations/AAI31298176