Background: Osteosarcoma (OS) is a highly aggressive primary bone malignancy, for which treatment and prognosis remain unchanged for three decades. This study aimed to delineate molecular subtypes of early-stage, treatment-naive OS to improve molecular stratification and guide precise therapeutic strategies. Methods: The top 10% of the most variable genes of 102 OS tumour tissues were used in NMF to classify the tumours into three subtypes, S1, S2, and S3. Differential expression analysis was carried out using DESeq2 and functional enrichment analysis using clusterProfiler and CancerHallmarks. Tumor microenvironment composition was estimated through ESTIMATE and CIBERSORT, and drug sensitivity prediction using OncoPredict. NMF was also applied to RNA-seq data from 19 osteosarcoma cell lines and DEGs between the groups were correlated with the DEGs from patients' subtypes. SAOS-2, representing S1, was used in the viability assays to determine the effects of hesperidin, alone or in combination with doxorubicin and 5-fluorouracil. Results: S1 was enriched in cell cycle regulation, vesicular transport, and RNA metabolism with high tumor purity and high resistance to chemotherapeutics. S2 and S3 were enriched in extracellular matrix organization and protein translation pathways, respectively. KIF20A was significantly overexpressed gene indicating a poor prognosis of S1. AURKB, a hesperidin target, was found as an important gene for the pathogenesis of S1. Hesperidin significantly reduced SAOS-2 cell viability and enhanced the efficacy of doxorubicin. Conclusions: Our results offer a strong molecular basis for OS subclassification, highlighting subtype-specific mechanisms of tumour progression and chemoresistance. Our findings support the prospect of customized therapeutic approaches, with hesperidin being a promising co-adjuvant to improve the prognosis of high-risk OS patients.