From 9000 Clues to a Clinical Test: Mass Spectrometry-Driven Lipidomics for Early Cancer Detection

Ovarian cancer is one of the deadliest cancers affecting women, largely because it is so rarely caught early. Its symptoms are vague, and existing blood tests often raise false alarms. Denver-based biotech company AOA Dx is aiming to change that with a "liquid biopsy," a blood test that looks for signs of cancer by analyzing several types of molecules at once, including lipids, metabolites, and proteins, and then uses machine learning to find patterns linked to the disease.¹ The company recently presented new data at the MSACL and EDCC conferences showing that its lead test, Omira™ OC, has progressed from a broad search through more than 9000 molecular clues to a targeted panel of fewer than 10. In an independent group of patients, the test distinguished ovarian cancer from non-cancerous conditions with similar symptoms with an AUC of 0.97, and early-stage disease with an AUC of 0.96. The findings also suggest the underlying biology reaches beyond ovarian cancer. A targeted lipid panel detected early-stage endometrial and cervical cancers with sensitivities of 95.7% and 94.4%, respectively, at 90% specificity, and early studies in other cancer types are showing encouraging signals.

High-resolution mass spectrometry, the analytical workhorse of modern lipidomics and metabolomics, is central to the company's discovery work. A key part of AOA Dx's strategy is designing Omira OC to run on the mass spectrometry and immunoassay systems that high-throughput clinical laboratories already use, offering a path to widespread adoption without the need for entirely new infrastructure.

LCGC International spoke to Abby McElhinny, Chief Scientific Officer of AOA Dx, about how the team turned complex multi-omics data into a reproducible clinical assay, the role of separation science and mass spectrometry in translating biomarker discovery into routine testing, and what these results could mean for the future of early cancer detection.

Omira OC began with more than 9000 candidate features and ended with a panel of fewer than 10. What strategy did your team use to narrow it down, and how did you balance diagnostic performance against what would be practical to run in a clinical laboratory?

From the beginning, our goal was not just to identify biomarkers associated with ovarian cancer, but to identify those that could ultimately be translated into a reliable, clinically practical diagnostic test.We used a combination of statistical and machine-learning approaches to analyze multiple cohorts of ovarian cancer and symptomatic non-cancer specimens, drawing from academic collaborators and our own prospectively collected clinical biobank. We specifically prioritized features that demonstrated consistent patterns across these independent cohorts, because reproducibility is essential for a biomarker to have clinical value.

Throughout this process, we evaluated more than diagnostic performance alone. Each candidate biomarker also had to demonstrate analytical precision, stability across different pre-analytical conditions, and reproducible predictive capabilities.

By systematically narrowing more than 9000 candidate features to a pool of fewer than 100 that maintained strong clinical performance, we established a foundation for further optimization. We then continued refining the panel as we transitioned to our clinical mass spectrometry platform, ultimately reducing it to fewer than 10 features.

The significance is that we've been able to dramatically simplify the assay while preserving the diagnostic performance needed for its intended clinical application. That translation from complex discovery science to a targeted, practical test is one of the most important steps toward bringing a new diagnostic into clinical use.

The platform combines mass spectrometry (MS) with immunoassay data. What does each one contribute that the other cannot, and how do you integrate the two kinds of measurements into a single result?

One of our most important findings has been that combining different types of biological information can provide a more powerful signal of ovarian cancer than relying on a single biomarker class.MS allows us to measure lipids and metabolites, while immunoassays measure proteins. Both platforms generate quantitative data, essentially measuring the concentration of specific features in the blood, which allows us to integrate these measurements using machine-learning approaches.

Some of these biomarkers increase in the presence of cancer, while others decrease compared with symptomatic women who do not have cancer. Our machine-learning classifier analyzes these combined patterns to calculate a probability of disease for each specimen. We then apply a defined decision threshold to generate a binary result indicating whether a cancer-associated signal has been detected and further clinical evaluation is warranted.

What we consistently observed was that proteins alone and lipids alone could distinguish ovarian cancer from non-cancer conditions. But combining the right proteins and lipids improved the sensitivity needed to detect Stage I and II ovarian cancer, even within the biologically complex population of symptomatic women.This is particularly important because early-stage ovarian cancer is where better diagnostic tools could make a meaningful difference, yet it remains one of the most challenging stages to detect.

Lipids appear to play a central role, particularly in the endometrial and cervical cancer results. What makes lipid metabolism such a strong source of cancer-associated signals, and what are the main analytical challenges in measuring lipids reproducibly in blood?

Lipids are involved in virtually every hallmark of cancer, something researchers have recognized for decades. They play essential roles in inflammation, cell survival and death, and the metabolic reprogramming that occurs as cancer develops and interacts with its surrounding environment.What makes lipids particularly interesting as biomarkers is that changes in these biological processes can produce measurable shifts in lipid profiles in the blood. Using the high resolution of mass spectrometry, we can quantify changes in specific lipids and observe how their profiles shift across different disease states.

The challenge is that lipids are chemically complex, and accurately identifying and measuring individual lipid species requires sophisticated analytical methods. Recent advances in lipid annotation, combined with our own work in identifying specific lipids from mass spectrometry data, have helped us address these challenges.By integrating these analytical advances with our computational biology expertise, we can evaluate which lipid changes are not only associated with cancer but also have reproducible predictive value.

This is what has allowed us to move beyond simply observing metabolic differences in cancer and begin translating those differences into a targeted diagnostic assay.

Moving from untargeted discovery to a targeted, clinical-grade assay is a big step. What changed in your liquid chromatography-mass spectrometry (LC-MS) workflow along the way, for example in sample preparation, chromatography, or acquisition mode?

One of the most significant milestones in our development has been moving from broad biomarker discovery to a targeted assay designed for clinical laboratory use.

We began with the orbital ion trap mass spectrometry platform, developing a novel method to identify as many lipid classes and subclasses as possible. Because lipids are highly abundant and diverse in blood, our initial challenge was distinguishing the features associated with ovarian cancer from the thousands of other signals that contribute biological noise.Once we established a reproducible cancer-associated lipid profile, we transitioned to a triple quadrupole (QqQ) mass spectrometry platform, which is well suited for targeted quantitative analysis in clinical laboratories. We then optimized the assay for analytical precision and accuracy.

An important part of this process was custom-synthesizing molecular standards of known concentration for each lipid, allowing us to strengthen confidence in lipid identity and develop a quantitative assay.

Finally, we selected the best-performing machine-learning model and optimized the decision threshold to support robust classification.The achievement here is not simply that we identified biomarkers associated with ovarian cancer. It's that we've taken a complex discovery signature and translated it into a much smaller, quantitative assay designed for the precision and reproducibility required in clinical diagnostics.

Existing ovarian cancer biomarkers are known to produce false positives in certain benign conditions. How did you test Omira OC against those conditions, and what do you think explains its higher specificity?

One of the challenges with existing ovarian cancer biomarkers is that benign conditions can produce overlapping signals, making it difficult to distinguish cancer from other causes of symptoms.We designed our approach specifically with this challenge in mind. From the earliest stages of discovery, we intentionally included specimens from symptomatic women with non-cancerous conditions, including benign gynecological and gastrointestinal diseases. These women have served as our primary control population throughout feature selection, model training, and testing.

This distinction is important. Rather than developing a test that simply differentiates women with ovarian cancer from healthy individuals, we have focused on distinguishing ovarian cancer from the conditions that can present with similar symptoms in real clinical settings.By developing and optimizing our assay against this complex population, we've worked to identify biomarker patterns that are more specific to ovarian cancer, while balancing sensitivity and specificity.Our goal has always been to build a test that performs in the population where the diagnostic uncertainty exists.

The assay was evaluated on an independent hold-out patient set. How was that set built, and what further validation will be needed before the test is used in routine care?

Our training and testing strategy has been designed around the intended clinical use of the assay: detecting ovarian cancer in symptomatic women, including those with benign conditions that can make diagnosis challenging.We use specimens from symptomatic women without cancer as our control population and enrich our datasets with early-stage ovarian cancer specimens obtained from multiple sources. This allows us to evaluate performance in the population we ultimately intend to serve, with particular emphasis on Stage I and II disease.

We also design our training and testing sets to provide the statistical precision needed to assess our target performance metrics, including performance around the assay's decision threshold.

We are now progressing through the next critical steps: analytical validation, final model lock, and clinical validation through our fully enrolled prospective study, ahead of our planned laboratory-developed test (LDT) launch in 2027.The focus now is on demonstrating that the performance we've observed during development can be reproduced with a finalized assay in the intended clinical population. That's an essential step in translating promising research findings into a test that clinicians can use with confidence.

Clinical laboratories vary in instruments, columns, and operators. How have you designed the assay to give consistent results across different laboratories and mass spectrometry platforms?

From the beginning, we've been thinking beyond the performance of the assay in a single laboratory. Our goal is to develop a test system that can ultimately be deployed more broadly while maintaining consistent analytical performance.To support that goal, we're designing OMIRA OC as a standardized, kitted solution that includes critical reagents, analytical methods, and the associated software pipeline.

Our initial launch will be as a laboratory-developed test in our Denver laboratory. However, our longer-term strategy is to decentralize testing by enabling additional clinical laboratories to implement the assay.By developing the system with that future deployment in mind, we can standardize key elements of the workflow, provide laboratories with the necessary reagents and software methods, and establish training procedures to support reproducible results.This is an important part of making advanced multi-omics diagnostics more scalable: designing not just for scientific performance, but for how a test can ultimately be implemented in real clinical laboratories.

Preanalytical factors such as how samples are collected, handled, and stored can strongly affect lipid and metabolite levels. How are you controlling for these variables?

For a blood-based diagnostic to be clinically useful, its performance must be robust to the variability that naturally occurs in real-world sample collection, transportation, and storage.We've evaluated more than 3500 specimens obtained from a variety of sources, with different storage durations and conditions. Incorporating this diversity throughout discovery and clinical assay development has helped us identify features that remain reproducible across a range of pre-analytical variables.We've also conducted extensive robustness testing, deliberately stressing samples beyond typical time and temperature conditions to understand how those factors may affect our biomarkers.

While analysis of those studies is ongoing, our results to date have not shown a significant impact on our selected features across the conditions evaluated.That's an encouraging finding because it suggests our biomarkers may be resilient to some of the practical variability that clinical laboratories encounter every day. Demonstrating that robustness is an important part of developing a diagnostic that can perform reliably outside a tightly controlled research environment.

You're seeing related signals across ovarian, endometrial, and cervical cancers. Do these cancers share common metabolic pathways, and how will you tell them apart if the same biomarkers show up in more than one?

One of the most exciting findings from our recent work is that we're observing related metabolic signals across ovarian, endometrial, and cervical cancers. This suggests that some of the biological changes we're measuring may reflect processes shared across different cancer types.We believe these features may be involved in common biochemical pathways, including energy metabolism, membrane remodeling within the tumor microenvironment, and the regulation of apoptosis, or programmed cell death.

At the same time, as we expand our analyses across metabolites, lipids, and proteins, we're identifying features that do not overlap across all three cancers and may be more indicative of a particular cancer type.This is an important distinction: our platform can capture both shared biological signals associated with cancer and more disease-specific signatures.Understanding that balance is a major focus of our next phase of research. It will enable us to build targeted diagnostic models for individual cancers while leveraging the broader biological insights generated by our multi-omics platform.

You've mentioned encouraging early signals in cancers outside gynecologic oncology. Without getting ahead of the data, what kinds of cancers or disease states do you think this approach is best suited to, and what's next for the platform?

We believe the potential of our platform extends well beyond ovarian cancer and even gynecologic oncology.Lipids and metabolites are involved in fundamental biological processes across many diseases, and our early findings demonstrate that measuring changes in these pathways could provide valuable diagnostic information across multiple indications.We've already observed differences in metabolic profiles between cancers and benign gynecological conditions, including endometriosis and PCOS. We're also generating exciting early data in non-gynecologic cancers, including early-stage disease in cancers that are particularly difficult to diagnose.

While these findings are still early, they reinforce our belief that the underlying science has applications beyond any single disease.

Our immediate priority remains advancing OMIRA OC through clinical validation and toward launch, while continuing to investigate how our discovery and targeted platforms can be applied to additional cancers and disease states.

What began as an effort to develop a better diagnostic test for ovarian cancer is now revealing the potential of a much broader platform. Our long-term vision is to use these measurable biological changes to develop a new generation of targeted blood tests that can identify disease earlier, more accurately, and in ways that are practical for clinical laboratories to implement.

References

1. AOA Dx Builds Oncology Platform Across Three Gynecologic Cancers as Lead Ovarian Cancer Assay Advances Toward Commercialization. AOA Dx website.https://aoadx.com/builds_oncology_platform/ (accessed 2026-10-16)