Integrated analysis of single-cell RNA-seq and spatial transcriptomics to identify the lactylation-related protein TUBB2A as a potential biomarker for glioblastoma in cancer cells by machine learning.

Background: An increasing number of studies have revealed a link between lactylation and tumor initiation and progression. However, the specific impact of lactylation on inter-patient heterogeneity and recurrence in glioblastoma (GBM) remains to be further elucidated.

Methods: We employed functional enrichment algorithms, including AUCell and UCell, to assess lactylation activity in GBM cancer cells. Additionally, we introduced the interquartile range (IQR) method based on a set of lactylation-related genes (LRGs) to reevaluate the extent of lactylation production within the cancer population at the single-cell resolution. By reconstructing the spatial transcriptomics of hematoxylin and eosin (HE)-stained sections, we further evaluated the lactylation activity in GBM tissues. Subsequently, We employed machine learning algorithms to identify hub genes significantly associated with elevated lactylation levels in GBM. Finally, we experimentally validated the emulsification efficiency and quantified the expression levels of hub genes in human GBM samples.

Results: Our study innovatively demonstrated a markedly elevated global lactylation level in GBM and validated it as an independent prognostic factor for GBM. We established a prognostic gene model associated with emulsification in GBM. Furthermore, the machine learning-based model identified SSBP1, RPA3 and TUBB2A as potential biomarkers for GBM. Notably, the expression levels of these three hub genes and the lactylation level of TUBB2A in GBM tissues were significantly higher compared to those in normal tissues.

Conclusions: We propose and validate a IQR lactylation screening method that provides potential insights for GBM therapy and an effective framework for developing gene screening models applicable to other diseases and pathogenic mechanisms.