Copy Number Alterations in Mechanobiological Genes for Metastatic-Status Classification: A Multicancer Machine-Learning Analysis

Metastasis is the leading cause of cancer-related mortality, highlighting the need for reliable diagnostic markers and molecular targets that distinguish primary and metastatic tumors. Genes involved in actin cytoskeleton organization and cell mechanics regulate cell migration and invasion and may represent both markers and functional targets of metastatic progression. This study investigated whether copy number alterations (CNAs) in an eleven-gene mechanobiology-related panel (FSCN1, GSN, ACTN4, EZR, CFL1, MYH9, PFN1, TAGLN, ECM1, ANXA2, ANXA6) can distinguish primary from metastatic tumors across breast, pancreatic, and gastric cancers. Publicly available cBioPortal datasets were analyzed using supervised machine-learning algorithms trained on gene-level CNA data. Models trained on combined multicancer datasets showed improved classification performance, evaluated on a held-out breast cancer test set, compared with breast cancer-only training, with Random Forest achieving the highest accuracy (0.851) and F1-score (0.844). Feature importance analysis identified PFN1, ECM1, and FSCN1 as the strongest contributors to metastatic classification, with ANXA6 also showing consistent relevance; permutation importance confirmed PFN1, ANXA6, and FSCN1 in both models. These findings suggest that copy number alterations in mechanobiology-related genes capture genomic patterns associated with metastatic status and may serve as candidate diagnostic biomarkers, warranting further functional and expression-level validation to assess their potential as therapeutic targets.