Parkinson's disease (PD), a progressive neurological disorder, manifests substantial clinical heterogeneity. Its phenotypic spectrum encompasses diverse motor and nonmotor manifestations, indicating probable discrete disease subtypes. Accurate identification of these subtypes is essential for guiding clinical management and personalized interventions. Currently, the classification of PD subtypes primarily depends on clinical behavioral assessments. Nevertheless, additional data modalities contribute significantly to Parkinson's disease subtyping. This investigation implemented similarity network fusion (SNF) to integrate multimodal datasets—encompassing functional MRI, DNA methylation profiles, and clinical behavioral metrics—obtained from the Parkinson's Progression Markers Initiative (PPMI) repository. Initially, we constructed individual patient similarity networks for each data modality, which were subsequently fused via the SNF method to generate a more comprehensive patient similarity network. Therefore, we applied the spectral clustering technique and successfully categorized PD patients into two distinct subtypes. Notably, significant differences in symptom severity and functional connectivity strength were observed between the two subtypes. Subtype 1 demonstrated markedly greater motor and nonmotor impairments, accompanied by a general reduction in functional connectivity levels relative to healthy controls. In contrast, compared with the control group, the Subtype 2 group presented with less severe symptoms and increased functional connectivity strength. The present work thus introduces an innovative analytical framework for the use of functional magnetic resonance imaging (fMRI) data to facilitate the clinical subtyping of Parkinson's disease.