Abstract Ovarian cancer (OC) is often diagnosed at late stages due to a lack of robust diagnostic tools, resulting in one of the highest mortality rates of any gynecologic malignancy. Dramatically altered lipidomic signatures have been observed in OC serum, but barriers exist in translating findings from discovery mass spectrometry (MS) applications to a targeted, clinical diagnostic assay. Discovery lipidomics experiments are challenged by the magnitude of features detected paired with the structural diversity of lipids. This work highlights the importance of feature characterization and identification to discover novel, robust biomarkers for the detection of early-stage OC.Serum samples representing individuals experiencing symptoms of OC as well as OC patients across all stages and a range of subtypes were obtained from University of Colorado, University of Manchester, and commercial vendors . Multi-omic analysis was performed using untargeted and targeted lipidomics and a panel of immunoassay protein biomarkers (CA125, HE4, MUC1, FOLR1). Discovery lipidomics used LCMS/MS with data-dependent acquisition (ddMS2). Data were processed for feature alignment, deconvolution, background rejection, identification, and statistical analysis. Feature exclusion included: (1) no library ID, (2) present in 10% samples, (3) exogenous origin, (4) high technical variability, and (5) not detected. Machine learning (ROC/AUC analysis) and univariate analyses identified features for targeted development. MRM transitions were generated from ddMS2 or targeted MS2 spectra, and targeted MRM methods were built. MRM features were experimentally verified in each polarity. Features were subjected to custom algorithm-informed internal standard normalization prior to inclusion in the model.Our published proof-of-concept multi-omic model reproducibly detects OC with AUCs of 92% (95% CI: 87%-95%) for OC v. controls and 88% (95% CI: 83%-93%) for early-stage OC. We then transferred the discovery lipidomics method to a targeted MRM assay, improving precision and analytical performance, while retaining directionality and/or significance in 82.5% features. Updated multi-omic models using targeted data show reproducible performance with AUCs 90% in an independent cohort of serum samples consistent with the discovery proof-of-concept multi-omic model (lipids + proteins). Here, we describe a workflow for translating discovery lipidomics into a targeted MRM method for a clinical diagnostic assay. Combining feature filtering, statistical analysis, machine learning, and experimental validation across cohorts, we identified a collection of robust lipid biomarkers that reproducibly detect OC. These features, combined with protein biomarkers, are being further developed into a multi-omic assay designed to detect OC earlier in the symptomatic population. Citation Format: Rachel Culp-Hill, Charles M. Nichols, Yu Han, Brendan M. Giles, Moisés Zapata, Mattie Goldberg, Robert A. Law, Enkhtuya Radnaa, Shannon Kilkenny, Maria Wong, Connor Hansen, Vuna S. Fa, Cory Bystrom, Liang Zhao, Kim Ekroos, Abigail McElhinny. Translating to targeted: Bridging discovery lipidomics to multi-omic clinical diagnostic application in ovarian cancer detection abstract. In: Proceedings of the American Association for Cancer Research Annual Meeting 2026; Part 1 (Regular Abstracts); 2026 Apr 17-22; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2026;86(7 Suppl):Abstract nr 2547.
Culp-Hill et al. (Fri,) studied this question.