Multi-modal data integration uncovers predictive biomarkers for ovarian cancer

Precision oncology has long promised a future in which each patient’s tumor is read like a molecular instruction manual, with treatments selected to match the specific genetic faults driving the disease. Yet the promise has repeatedly stumbled in clinical trials, where aberration-guided therapies have too often failed to deliver the expected benefit. A new study published in Genome Medicine suggests a major reason why: a startling proportion of the genetic variants that clinicians treat as actionable are, in fact, false leads. By systematically integrating whole-genome and transcriptome sequencing data from hundreds of patients with ovarian high-grade serous carcinoma, an international research consortium has developed a workflow capable of separating functionally credible, treatment-guiding aberrations from inert molecular noise, and in doing so has exposed both the scale of the false-positive problem and a set of genuine therapeutic vulnerabilities in one of the deadliest gynecologic malignancies.

The research draws on the DECIDER study, an observational cohort of 335 patients with ovarian high-grade serous carcinoma, or HGSC, the aggressive histologic subtype responsible for the vast majority of ovarian cancer deaths. Tumor samples were collected from multiple anatomical cancer sites as part of each patient’s standard clinical care, ensuring that the molecular portrait captured by the team reflected the true heterogeneity of the disease rather than the narrow view offered by a single biopsy. DNA and RNA were extracted together from snap-frozen tumor samples and subjected to whole-genome and transcriptome sequencing, a dual approach that allows researchers to see not only which genes are altered but also whether those alterations leave detectable traces in the tumor’s transcriptional output.

Processing this torrent of genomic data required a substantial bioinformatics infrastructure. Short somatic changes were detected and validated with the Anduril 2 pipeline, while structural alterations—large-scale rearrangements of the genome that are particularly characteristic of HGSC—were assessed with the HMW toolkit and the nf-core/rnafusion pipeline. This multi-modal architecture is the technical heart of the study. Rather than treating each class of genomic event in isolation, the workflow cross-references DNA-level aberrations against RNA-level expression signatures, asking a deceptively simple question of every candidate variant: does it actually do anything to the tumor? Only aberrations that pass this functional credibility filter are promoted to the status of potential predictive biomarkers.

The results were striking. Using an agnostic, integrated omics analysis, the team identified clinically relevant alterations classified as Tier II or Tier III on the ESMO Scale for Clinical Actionability of Molecular Targets—ESCAT, the standard framework for ranking the evidence supporting biomarker-guided therapy—in more than 40 percent of the patients. On its face, that figure represents a substantial pool of women who might benefit from targeted agents. But the more provocative finding lurked beneath the surface: 58 percent of all nominally pathogenic variants turned out to be false positives. In other words, well over half of the genetic changes that conventional annotation pipelines would flag as potentially actionable appear to have no functional consequence in the tumor whatsoever.

The implications of that false-positive rate extend far beyond academic bookkeeping. A patient whose tumor harbors an inert hotspot mutation in a gene like BRAF or EGFR—both exemplified in the study’s supplementary case analyses—might be enrolled in a biomarker-matched clinical trial or offered an off-label targeted drug, receiving a treatment that has no chance of working while forfeiting time and tolerability on regimens that might help. At the level of entire trials, phantom biomarkers dilute study populations with patients whose tumors do not genuinely depend on the targeted pathway, a well-recognized mechanism by which promising precision oncology trials fail to meet their endpoints. The ability to discriminate treatment-guiding aberrations from false positives, the authors argue, could directly improve the design and interpretation of biomarker-guided trials.

How, then, does the integrated workflow tell a real driver from a decoy? The study found that credible aberrations shared a consistent evolutionary and transcriptional signature. They were predominantly clonal, meaning the alteration was present in essentially all tumor cells rather than confined to a minor subpopulation, and they were detected consistently across different anatomical sites sampled from the same patient. Crucially, they were preserved from diagnosis to relapse, indicating that these alterations are established early during tumor evolution and maintained throughout the disease course—exactly the profile expected of alterations on which the tumor has become fundamentally dependent, and exactly the profile that makes them reliable targets for therapy.

The most recurrent actionable event identified through this filtering process was deficiency of NF1, a tumor suppressor gene whose product normally acts as a brake on RAS-family signaling. NF1 loss left a robust and characteristic transcriptional footprint in the patients’ tumors, satisfying the workflow’s requirement that a credible aberration manifest at the RNA level as well as in the genome. More importantly, the finding translated directly into a therapeutic hypothesis. When tumor cells from NF1-deficient patients were grown as patient-derived organoids—three-dimensional cultures that recapitulate key features of the original tumor—and subjected to a drug screen combining targeted agents with chemotherapy, they displayed marked sensitivity to inhibitors of KRAS and MEK, two downstream components of the very signaling pathway that NF1 normally restrains.

This NF1–KRAS–MEK axis represents a genuinely unexploited vulnerability in HGSC, a disease in which validated targeted options beyond PARP inhibitors for homologous recombination–deficient tumors remain scarce. The organoid-based drug screen provides a functional validation layer that pure genomic correlation studies lack: the sensitivity was not merely associated with the biomarker but demonstrated experimentally in living tumor material derived from the patients themselves. Drug synergy analyses in the organoid platform further suggest that combinations building on this axis may warrant formal clinical investigation, informing the biomarker-guided trial designs that the authors explicitly cite as one of the study’s aims.

The methodological achievement should not be understated. Whole-genome sequencing at scale generates terabytes of raw data per patient, and the decision to process DNA and RNA from the same snap-frozen samples through harmonized pipelines—Anduril 2 for short variants, HMW and nf-core/rnafusion for structural events—minimizes the batch effects and sample mismatches that plague many retrospective multi-omics studies. The agnostic nature of the analysis is equally important: rather than preselecting a panel of known hotspots, the workflow interrogates all classes of genomic events, from single-nucleotide variants to large structural rearrangements, and lets functional evidence decide which of them matter.

For patients with HGSC, the clinical stakes are enormous. The disease is typically diagnosed at an advanced stage, initially responds to platinum-based chemotherapy and, where appropriate, PARP inhibitor maintenance, but relapses are nearly universal and each successive line of therapy buys less time. If the integrated workflow can be operationalized in routine molecular diagnostics, the more than 40 percent of patients carrying credible ESCAT Tier II–III alterations could be directed toward targeted trials with biomarkers they can trust, while the majority carrying only false-positive variants are spared futile exposure to mismatched drugs. The study’s authors, a consortium spanning the University of Helsinki, the University of Copenhagen, the Institut Pasteur, Heidelberg University, the University of Modena and Reggio Emilia, and other centers, conducted the work with ethical approval and written informed consent from all participants, and note that the platform also revealed unexplored therapeutic vulnerabilities beyond NF1 deficiency.

The broader message resonates across precision oncology as a whole. As sequencing becomes cheaper and biomarker panels more comprehensive, the bottleneck is no longer detection but interpretation—knowing which of the hundreds of alterations catalogued in a tumor report are worth acting upon. This study demonstrates that cross-referencing the genome against the transcriptome, validating candidates in patient-derived models, and demanding clonality and persistence as evolutionary credentials offers a concrete, technically mature strategy for making that distinction. In a field where a single wrong biomarker call can cost a patient precious months, the ability to discard more than half of the nominally pathogenic variants as false positives is not a cautionary footnote. It may be the difference between precision oncology that works and precision oncology that merely looks like it does.