A Novel Classifier Based on Urinary Proteomics for Distinguishing Between Benign and Malignant Ovarian Tumors

Maowei Ni; Jie Zhou; Zhihui Zhu; Jingtao Yuan; Wangang Gong; Jianqing Zhu; Zhiguo Zheng; Huajun Zhao

doi:10.3389/fcell.2021.712196

A Novel Classifier Based on Urinary Proteomics for Distinguishing Between Benign and Malignant Ovarian Tumors

Front Cell Dev Biol. 2021 Aug 30:9:712196. doi: 10.3389/fcell.2021.712196. eCollection 2021.

Authors

Maowei Ni^{1

2

3}, Jie Zhou^{4

5}, Zhihui Zhu¹, Jingtao Yuan¹, Wangang Gong^{2

3}, Jianqing Zhu^{2

3}, Zhiguo Zheng^{2

3}, Huajun Zhao¹

Affiliations

¹ School of Pharmaceutical Sciences, Zhejiang Chinese Medical University, Hangzhou, China.
² The Cancer Hospital of the University of Chinese Academy of Sciences (Zhejiang Cancer Hospital), Hangzhou, China.
³ Institute of Basic Medicine and Cancer (IBMC), Chinese Academy of Sciences, Hangzhou, China.
⁴ Department of Physiology, Zhejiang Chinese Medical University, Hangzhou, China.
⁵ Tongde Hospital of Zhejiang Province, Zhejiang Academy of Traditional Chinese Medicine, Hangzhou, China.

Abstract

Background: Preoperative differentiation of benign and malignant tumor types is critical for providing individualized treatment interventions to improve prognosis of patients with ovarian cancer. High-throughput proteomics analysis of urine samples was performed to identify reliable and non-invasive biomarkers that could effectively discriminate between the two ovarian tumor types.

Methods: In total, 132 urine samples from 73 malignant and 59 benign cases of ovarian carcinoma were divided into C1 (training and test datasets) and C2 (validation dataset) cohorts. Mass spectrometry (MS) data of all samples were acquired in data-independent acquisition (DIA) mode with an Orbitrap mass spectrometer and analyzed using DIA-NN software. The generated classifier was trained with Random Forest algorithm from the training dataset and validated in the test and validation datasets. Serum CA125 and HE4 levels were additionally determined in all patients. Finally, classification accuracy of the classifier, serum CA125 and serum HE4 in all samples were evaluated and plotted via receiver operating characteristic (ROC) analysis.

Results: In total, 2,199 proteins were quantified and 69 identified with differential expression in benign and malignant groups of the C1 cohort. A classifier incorporating five proteins (WFDC2, PTMA, PVRL4, FIBA, and PVRL2) was trained and validated in this study. Evaluation of the performance of the classifier revealed AUC values of 0.970 and 0.952 in the test and validation datasets, respectively. In all 132 patients, AUCs of 0.966, 0.947, and 0.979 were achieved with the classifier, serum CA125, and serum HE4, respectively. Among eight patients with early stage malignancy, 7, 6, and 4 were accurately diagnosed based on classifier, serum CA125, and serum HE4, respectively.

Conclusion: The novel classifier incorporating a urinary protein panel presents a promising non-invasive diagnostic biomarker for classifying benign and malignant ovarian tumors.

Keywords: machine learning; mass spectrometry; non-invasive biomarkers; ovarian cancer; urinary proteomics.