Background: The association between microbiome and host disease has been documented in oral cancer, one of the leading cancers worldwide. Huge efforts are made to use the profile of oral microbiome and distinct signature species as markers to distinguish oral cancer patients from healthy individuals. The previous results, however, remain inconclusive. The assembly mechanisms of oral microbiome and their response to changes in oral carcinogenesis also have not been characterized. Here, using 16S rRNA gene amplicon sequencing and in-silico function prediction approaches, we analyzed the saliva microbiome in cohorts of orally healthy, oral verrucous hyperplasia, and oral cancer at taxon and function levels, and compared their corresponding predictive performance of oral cancer using machine learning algorithms.Results: Analyses of diversity and phylogenetic profiles of bacterial communities in saliva showed that microbiome dysbiosis was significantly linked to oral health status. As oral health deteriorated, the number of core species (>75% prevalence) as a percentage of overall species richness declined. In line with the null model-based analysis, taxonomic and functional assemblies of saliva microbiomes were primarily governed by the stochastic processes. Correspondingly, the quantitative assessment of partitioned beta-diversity suggested extremely high species turnover but low function turnover, revealing a functional redundancy of the oral ecosystem. Functional beta-diversity in salvia microbiome shifted from turnover to nestedness during carcinogenesis of oral verrucous hyperplasia, but this pattern was not observed at the taxon level. Moreover, using both taxon and function data as training features for machine learning-aided prediction on host health status supports a superior predictive performance when using functional profiling. Similar results were also obtained and validated using publicly accessible data.Conclusions: Our results suggest that altered oral bacterial communities are highly associated with carcinogenesis of oral verrucous hyperplasia. Partly owing to high taxonomic turnover and stochastic assembly processes of the oral ecosystem, discovering oral microbial consortia as universal biomarkers for oral cancer may prove difficult and arduous. Functional profiles are relatively stable and evolve a nestedness pattern during oral carcinogenesis, serving as a new benchmark to study the interplay of the oral microbiome and host health in the future.