Abstract B077: Unsupervised computational characterization of circulating microRNA networks defines plasma-based tumor phenotypes

Abstract Background: Prostate cancer presents significant clinical heterogeneity, making it difficult to predict tumor aggressiveness and guide therapy using tissue biopsies alone. Circulating microRNAs (miRNAs) in plasma represent promising, minimally invasive biomarkers that may reflect the molecular phenotype of the primary tumor. Methods: Using NanoString technology, we analyzed paired plasma and primary tumor samples from 24 patients with diverse prostate cancer subtypes to evaluate whether plasma miRNA expression patterns could stratify tumor phenotypes. Absolute global normalization and z-score transformation were applied to the miRNA sequencing data to minimize technical and biological variability. Dimensionality reduction using principal component analysis (PCA) was employed to uncover intrinsic structure and identify pattern-defining miRNAs. A unique pattern-based identity was generated for each primary and corresponding plasma sample, allowing comparison between their spatial distributions in reduced-dimension space. Results: Optimized PCA revealed distinct clustering of plasma miRNA profiles that mirrored the groupings observed for the corresponding primary tumors. These shared expression patterns enabled discrimination between tumor subtypes using circulating miRNAs alone, despite minimal overlap in specific miRNA identities between plasma and tissue. This indicates that the relative expression relationships among miRNAs, rather than the presence of identical miRNAs, carry information reflective of the underlying tumor phenotype. The approach leverages plasma miRNA expression as a surrogate signature for tumor characterization without requiring matched expression concordance between compartments. Conclusions: Our findings demonstrate that dimensionality-reduction analysis of plasma miRNA expression can stratify prostate cancer phenotypes and reflect tissue-level molecular diversity. This unsupervised, pattern-based framework provides a foundation for non-invasive tumor profiling and may enable future precision diagnostics using circulating miRNA signatures. Citation Format: Gobi Thillainadesan, Yutaka Amemiya, Robert Nam, Arun Seth. Unsupervised computational characterization of circulating microRNA networks defines plasma-based tumor phenotypes [abstract]. In: Proceedings of the AACR Special Conference in Cancer Research: Innovations in Prostate Cancer Research and Treatment; 2026 Jan 20-22; Philadelphia PA. Philadelphia (PA): AACR; Cancer Res 2026;86(2_Suppl):Abstract nr B077.