Integrative multi-omics refines the molecular subtypes of thyroid cancers and enhances cancer-progression prediction: a retrospective cohort study.

Current molecular classification model for thyroid cancer (TC), which relies on BRAF-RAS score genes has limited efficacy in differentiating follicular-patterned tumors from normal thyroid (NT) tissues and lack predictive power for disease progression. This study aimed to refine TC classification and develop predictive systems for tumor progression. A multi-omics analysis integrating transcriptomics (SNUH-mRNASeq) and proteomics (SNUH-TMT-Pro) from retrospectively collected human tissues representing NT and tumors with mutation profiles was conducted in a single center. A novel gene set, termed NPF genes, was identified and used to construct a decision-tree based classification system and a risk stratification model for tumor progression. The discriminatory potential of protein markers was validated through immunohistochemistry (SNUH-IHC) using tissue microarrays. External datasets (TCGA-THCA, SNUH-DIA-Pro, and CellDis-Pro) were employed to validate both the classification and risk stratification systems. Clustering based on NPF genes separated papillary thyroid cancer (PTC) with BRAFV600E mutation (PTC-B), follicular thyroid cancer (FTC) regardless of mutation status and NT, classifying them into BRAFV600E-like, RAS-like, and NT-like subtypes. The decision-tree based classification model with NPF genes demonstrated high accuracy (0.92, 95% CI 0.88-0.95) and Kappa statistics (0.88). IHC of three protein biomarkers (MATN2, FN1, and PLSCR4) confirmed the molecular subtypes in SNUH-IHC, with findings consistent across external datasets. Additionally, higher NP-score and NF-score predicted poorer prognosis in BRAFV600E-like and RAS-like TCs. The NPF gene set and classification model refine TC classification, improve diagnostic accuracy, and enable better risk stratification. These advancements offer a foundation for personalized therapeutic strategies in TC management.