Abstract
BACKGROUND: Transfer RNA-derived small RNAs (tsRNAs) represent an emerging class of regulatory molecules with potential as cancer biomarkers. However, their diagnostic utility and regulatory mechanisms in breast cancer remain poorly characterized. This study integrates machine learning algorithms with traditional molecular biology approaches to identify tsRNA-based diagnostic signatures and their downstream targets. METHODS: We analyzed miRNA-seq data from 103 matched tumor-normal pairs from TCGA-BRCA as the discovery cohort and GSE117452 as validation. tsRNA profiles were extracted using a custom bioinformatics pipeline. Random forest algorithm was employed to develop a diagnostic model. Correlation analysis and RNAhybrid were used to identify tsRNA-mRNA regulatory relationships. Comprehensive multi-omics analyses including survival, immune infiltration, drug sensitivity, and pathway enrichment were performed for identified targets. Functional validation was conducted in breast cancer cell lines. RESULTS: We identified 297 differentially expressed tsRNAs and developed a four-tsRNA signature (tRF-21-FSXMSL73E, tRF-20-XSXMSL73, tRF-23-FSXMSL730H, tRF-23-YJE76INB0J) achieving AUC of 0.98 in discovery and 0.82 in validation cohorts. tRF-21-FSXMSL73E showed strong correlation with FAM155B expression. Pan-cancer analysis revealed FAM155B overexpression in multiple malignancies with prognostic significance. FAM155B correlated with immune infiltration, drug resistance, and activation of oncogenic pathways. Functional studies confirmed FAM155B promotes breast cancer proliferation and migration. CONCLUSION: Our machine learning approach successfully identified a robust tsRNA diagnostic signature and uncovered the tsRNA-FAM155B regulatory axis as a novel therapeutic target. This integrated methodology provides a framework for accelerating biomarker discovery by combining computational prediction with traditional validation, advancing precision medicine in breast cancer.