The Malignant Feature-Based Prognostic Signature: A Machine Learning-Based Model Predicting Clinical Outcomes and Unveiling OPN3 as a Novel Therapeutic Target Candidate in Osteosarcoma

Osteosarcoma is the most common primary malignant bone tumor in children and adolescents, characterized by high invasiveness, early metastasis, and frequent recurrence, which are largely driven by genomic copy number variations (CNVs). However, existing prognostic models based on bulk RNA-seq are constrained by intratumoral heterogeneity and fail to reflect intrinsic malignant cell characteristics. Based on single-cell transcriptomic data, osteoblastic cells were stratified by CNV burden into high- and low-malignancy subpopulations, and malignant feature-related genes (MFGs) were identified. An integrative machine-learning algorithm was used to construct the malignant feature-based prognostic signature (MFPS) prognostic model, validated in independent cohorts. We then analyzed the MFPS-associated immune microenvironment, drug sensitivity, and biological mechanisms, and screened key genes. Finally, knockdown experiments in osteosarcoma cell lines assessed proliferation, cell cycle, migration, invasion, and transcriptomic changes. The high-grade malignant osteoblastic cell (HG-MOB) subpopulation exhibited high CNV burden, enhanced proliferation and intercellular communication, and low differentiation. The MFPS model, built from 11 MFGs, performed robustly across multiple cohorts and outperformed most reported models. High-MFPS patients had shorter overall survival and higher metastasis and recurrence risks. The high-MFPS group showed an immunosuppressive microenvironment, with reduced CD8⁺ T and NK cells, M2 macrophage enrichment, immunomodulatory molecule downregulation, and activation of mTORC1, hypoxia, and glycolysis. OPN3 was prioritized among MFGs. OPN3 knockdown modestly reduced proliferation but significantly arrested G₁/S transition and suppressed migration and invasion. RNA-seq revealed perturbations in PI3K-Akt, TNF, and cytokine-cytokine receptor pathways, with downregulation of angiogenesis- and hypoxia-related gene sets. This study identifies a CNV-defined HG-MOB subpopulation and establishes MFPS as a valuable prognostic tool reflecting tumor aggressiveness and immune heterogeneity. OPN3 is validated as a key driver of osteosarcoma malignancy, potentially via the PI3K-Akt axis and downstream angiogenic and metabolic pathways, representing a novel candidate therapeutic target.