Cognitive impairment (CI) is very common in patients with Parkinson’s Disease (PD) and progressively develops on a spectrum from mild cognitive impairment (PD-MCI) to full dementia (PDD). Identification of PD patients at risk of developing cognitive decline, therefore, is unmet need in the clinic to manage the disease. Previous studies reported that oral microbiota of PD patients was altered even at early stages and poor oral hygiene is associated with dementia. However, data from single modalities often unable to explain complex chronic diseases in the brain and cannot reliably predict the risk of disease progression. Here, we performed integrative metaproteogenomic characterization of salivary microbiota and tested the hypothesis that biological molecules of saliva and saliva microbiota dynamically shift in association with the progression of cognitive decline and harbor discriminatory key signatures across the spectrum of CI in PD. We recruited a cohort of 115 participants in a multi-center study and employed multi-omics factor analysis (MOFA) to integrate amplicon sequencing and metaproteomic analysis to identify signature taxa and proteins in saliva. Our baseline analyses revealed contrasting interplay between the genus Neisseria and Lactobacillus and Ligilactobacillus genera across the spectrum of CI. The group specific signature profiles enabled us to identify candidate biomarkers including 7 bacterial genera (Neisseria, Lactobacillus, Rothia, Ligilactobacillus, Alloprevotella, TM7x and Corynebacteirum) and 4 protein groups (40S ribosomal protein SA, 40S ribosomal protein S15, pyruvate, phosphate dikinase and bactericidal permeability-increasing protein) discriminating CI stages in PD (AUC 0.74-0.86). Our study describes compositional dynamics of saliva across the spectrum of CI in PD and paves the way for developing novel, non-invasive biomarker strategies to predict the risk of CI progression in PD.