In:
Cancer Research, American Association for Cancer Research (AACR), Vol. 83, No. 7_Supplement ( 2023-04-04), p. 5549-5549
Abstract:
Owing to a lack of disease-specific symptoms and effective screening tools, most ovarian cancers are diagnosed at an advanced stage, resulting in high recurrence and mortality rates. Recently, there has been an effort to diagnose ovarian cancer utilizing platelets, which are crucial in the immunology of oncogenesis. These transcriptome-wise techniques, however, had two significant drawbacks: utilizing a normalization method that can be altered depending on the content of the samples in the dataset and using an excessive number of features. We, therefore, present a method for assessing the existence of ovarian cancer or benign tumors by utilizing a fewer features with a sampling invariant normalization technique. First, we downloaded platelet transcriptome data of patients with ovarian cancer (n=28), benign ovarian tumor (n=17), and their healthy counterparts (n=204) from GEO (GSE158508 and GSE89843 - PRJNA353588 and PRJNA353588). Second, for external validation, we prospectively enrolled patients with ovarian cancer (n=4) and benign ovarian tumors (n=9), and healthy women (n=14) at Seoul National University Hospital (SNUH), collected their blood samples, and obtained platelet transcriptome data. The normalization was performed using commonly and invariantly existing 20 splice junctions in the isolated platelet of tumor and normal samples. Herein, 319 splice junctions were selected as features for the SVM model diagnosing the existence of an ovarian tumor. Feature selection and model development were conducted using a training set (n=152) and model performance was assessed from a separate test set (n=97). SVM classifier that utilized our splice junction-based biomarkers (20 for normalization, 319 for tumor classification) demonstrated 93.8% (71.7 - 98.9%) sensitivity, 100.0% (95.5 - 100.0%) specificity, and 1.0 of AUC in the test data set with a predetermined 0.5 cut-off value. Considering the difference of sequencing raw data between the open source and newly collected sample in terms of read length and read types (single-end vs paired-end), the SVM model from the open source was not suitable to the independent validation set. Thus, we train new SVM models using the same splice junctions with LOOCV in SNUH dataset. As the results, the model showed its diagnostic performance with 92.3% (66.7 - 98.6%) sensitivity and 92.9% (68.6 - 98.7%) specificity. We found novel splice junction-based biomarkers for the early detection or diagnosis of ovarian tumors. Although possible confounding factors such as ethnic variance may affect the performance of the models using these biomarkers, since the data preprocessing procedure is sampling invariant, the inclusion of samples from the same population would not affect our classification results. Furthermore, suggested biomarkers can be utilized during medical checkups of women without any symptoms to find women with ovarian tumors, whether benign or malignant. Citation Format: Eunyong Ahn, Se Ik Kim, Sungmin Park, Sarah Kim, Seung Jin Yang, Yeochan Kim, Dong Won Hwang, Heeyeon Kim, HyunA Jo, Untack Cho, Juwon Lee, Yong-Sang Song, TaeJin Ahn. Normalized platelet splicing junction count is a novel biomarker for diagnosis of ovarian tumors. [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2023; Part 1 (Regular and Invited Abstracts); 2023 Apr 14-19; Orlando, FL. Philadelphia (PA): AACR; Cancer Res 2023;83(7_Suppl):Abstract nr 5549.
Type of Medium:
Online Resource
ISSN:
1538-7445
DOI:
10.1158/1538-7445.AM2023-5549
Language:
English
Publisher:
American Association for Cancer Research (AACR)
Publication Date:
2023
detail.hit.zdb_id:
2036785-5
detail.hit.zdb_id:
1432-1
detail.hit.zdb_id:
410466-3
Permalink
