StemnesScoRe

StemnesScoRe estimates the stemness of glioma cancer cells at single-cell resolution using scATAC-seq data and machine learning techniques to quantify stemness scores.

Key Features:

• Single-cell stemness quantification: Calculates stemness scores for individual glioma cells using scATAC-seq peak regions and gene importance weights.
• Machine learning-based gene selection: Applies random forest, gradient boosting, and extreme gradient boosting algorithms to glioblastoma scRNA-seq data to identify genes most predictive of cellular states.
• Regulatory region integration: Maps scATAC-seq peaks to promoter and enhancer regions associated with state-predictive genes via overlapping region analysis.
• Glioma-specific modeling: Develops a dedicated model for glioblastoma stemness estimation using scATAC-seq data for the first time.

Scientific Applications:

• Glioblastoma heterogeneity analysis: Assesses inter- and intra-tumoral heterogeneity by quantifying stemness across diverse glioma cell populations.
• Cancer stem cell identification: Distinguishes cells with high stemness characteristics, such as mesenchymal-like stem cells, from neural-progenitor-, oligodendrocyte-progenitor-, and astrocyte-like cells.
• Therapy resistance research: Supports investigation of glioma cancer stem cell plasticity and resistance to multiple treatment modalities.

Methodology:

StemnesScoRe applies random forest, gradient boosting, and extreme gradient boosting to glioblastoma scRNA-seq data to identify genes associated with cellular states, then identifies promoter and enhancer regions for these genes. The method downloads scATAC-seq peaks and read counts per patient, identifies overlapping regions between single-cell peaks and machine learning-derived gene peaks, calculates read counts mapped to overlapping regions, and estimates stemness scores using overlapping regions combined with gene importance values predictive of glioblastoma cellular states.