Abstract 7P
Background
High-grade serous ovarian cancer, despite its high lethality, lacks reliable biomarkers for predicting poor prognosis, and Limited progression has been made in personalized treatment. Genomic profile-based targeted therapy has not met expectations, as genomic alterations alone do not exclusively determine cancer cell phenotypes. Protein expression critically influences cellular processes. Recognizing proteomic alterations is even more crucial. This study proposes a novel technique, utilizing statistical deviation and machine- learning to select protein factors determining ovarian cancer prognosis.
Methods
In advanced high-grade serous ovarian cancer patients, divided into two groups with very good (n=23) and poor prognoses (n=24), proteins were extracted from fresh frozen tissue and subjected to proximity extension assay (PEA). We explored a novel approach called AI-based machine learning to identify key proteins that could distinguish between groups with good and poor prognoses. Proteins were validated by immunohistochemistry (IHC) staining and cell proliferation assay, transwell migration assay, and Boyden chamber invasion assay.
Results
We explored a novel approach called AI-based machine learning to identify key proteins that could distinguish between groups with good and poor prognoses. By developing a model, we found that high levels of NPTN and PPM1A indicated a poor prognosis group, demonstrating remarkably high efficacy (Precision 0.857, Recall 0.818, F1-score 0.893). After IHC of NPTN and PPM1A in a tissue microarray (TMA), survival analysis showed that survival decreased when the expression was high. In vitro experiments with NPTN and PPM1A knockdown showed reduced cell proliferation, migration, and invasion.
Conclusions
Our results suggest that it is feasible to select factors with significant differences between prognostic groups, particularly those that are amenable to clustering based on identified proteins. The research highlights the potential of proteomic markers to guide personalized therapeutic strategies to improve patient outcomes.
Legal entity responsible for the study
The authors.
Funding
Has not received any funding.
Disclosure
All authors have declared no conflicts of interest.