New 9-Gene Classifier May More Accurately Predict Metastasis Across STS, Other Cancers

A newly developed gene classifier may significantly improve clinicians’ ability to predict which soft-tissue sarcoma (STS) patients are most likely to develop distant metastases. The same tool also appeared to work across several other major cancer types, suggesting it may be tapping into core biological mechanisms that govern metastatic progression across multiple malignancies.

Publishing their findings in Cancer Treatment and Research Communications, researchers showed that their prediction model leveraging 9 genes related to metastasis in SIS accurately stratified patients into metastasis risk, outperforming many existing prognostic gene signatures, including CINSARC, which, based on 67 genes, divides patients into 2 risk groups.2

“One of the major implications of developing prognostic prediction models is to provide cancer patients and health workers with information that will be useful in making treatment decisions,” wrote the researchers.1 “Although attempts to identify common genetic characteristics in STS with very diverse histology and to use these characteristics to predict patient prognosis are already underway at various research institutes, there is currently no [gene expression profiles] test for clinical diagnosis of STS.”

To benchmark performance, the researchers compared the model’s accuracy with 5 widely used prognostic signatures. | Image credit: vladimircaribb - stock.adobe.com

Against that backdrop, the researchers analyzed thousands of tumor samples from public genomic databases to produce a machine learning–driven model consisting of genes consistently associated with metastasis-free survival.

The group identified and leveraged 34 genes that showed strong associations across datasets and then used machine learning to determine whether a much smaller subset of genes could be used to classify patients. Testing more than a million gene-combination patterns through iterative stratified cross-validation, the team identified the optimal combination of TNXB, ABCA8, ACTN1, EIF4EBP1, PVR, CLIC4, STAU2, ATAD2, and TBC1D31.

Producing the highest accuracy, the 9-gene set produced became the basis of the final classifier. When applied to multiple STS datasets, the model consistently separated patients into low-risk and high-risk groups, with strong statistical significance across nearly all cohorts.

The classifier’s performance extended beyond sarcoma, showing the ability to distinguish between favorable and poor prognoses in breast cancer data sets, with high-risk groups demonstrating sharply higher rates of distant metastasis, especially to the lungs and brain. Notably, the tool was able to identify which breast cancer subgroups were likely to benefit from adjuvant chemotherapy, which could help clinicians avoid unnecessary toxicity in patients unlikely to benefit.

The classifier also performed well in kidney clear cell carcinoma and uveal melanoma, 2 cancers where metastasis strongly influences survival. Across these datasets, the 9-gene model continued to assign patients into prognostic groups with distinctive metastatic patterns and disease-specific survival outcomes.

To benchmark performance, the researchers compared the model’s accuracy with 5 widely used prognostic signatures. In nearly all STS datasets, the 9-gene classifier achieved higher or more stable area under the curve (AUC) scores, surpassing CINSARC in 3 of 4 major datasets. Its predictive stability across diverse cancers also exceeded that of several other signatures, except for Vijver’s 70-gene breast cancer signature, which remained one of the strongest performers in breast cancer but fared less well in sarcoma and uveal melanoma.

While promising, the researchers acknowledged limitations of their model, including its poor performance in pediatric rhabdomyosarcoma, suggesting age-specific or subtype-specific biology may require tailored approaches. Additionally, most datasets included fresh-frozen tumor samples; clinical translation will require validation using formalin-fixed, paraffin-embedded tissue commonly collected in diagnostic workflows.

References

1. Tanabe A, Ndzinu J, Sahara H. A novel 9-gene classifier for predicting distant metastasis of soft-tissue sarcoma and multiple malignancies. Clin Treat Res Commun. Published online November 28, 2025. doi:10.1016/j.ctarc.2025.101046

2. Callegaro D, Tinè G, Oppong FB, et al. CINSARRC and sarculator in patients with primary retroperitoneal sarcoma: a combined analysis of single-institution data and the EORTC-STBSG-62092 trial (STRASS). Clin Cancer Res. 2025;31(15):3239-3249 doi:10.1158/1078-0432.CCR-25-0099