Journal: Computational and Structural Biotechnology Journal

Article Title: QeITH: Quantifies Tumor Ecosystem Heterogeneity to Predict Cancer Progression and Treatment Benefit

doi: 10.34133/csbj.0061

Figure Lengend Snippet: Robustness and benchmarking of QeITH. (A) Evaluation of QeITH performance in distinguishing immunotherapy responders from nonresponders using matched SKCM bulk and single-cell data. Left: Scatterplot showing strong Spearman correlation between ITH scores derived from bulk RNA-seq and cell composition-based ITH scores from single-cell data. Right: Boxplots with jittered points showing QeITH and DEPTH2 scores in responders versus nonresponders. (B) Pan-cancer comparison of QeITH and DEPTH2 scores between tumor and normal tissues in pseudobulk samples derived from scRNA-seq data. ITH scores were significantly elevated in tumors compared to normal tissues in pan-cancer analysis and validated in lung, pancreatic, and gastric cancers individually. DEPTH2 failed to detect significant tumor-normal differences in pan-cancer, pancreatic, or gastric cancer, and unexpectedly showed higher scores in normal lung tissues compared to tumors. Dot plots show individual sample values; each dot represents one sample. White diamonds indicate median values for each group. For the remaining panels, half-violins (right side) show density distributions; boxplots with overlaid line segments show median, IQR, and individual sample values (each line segment represents a single sample). (C) Stability of QeITH and ROGUE across clustering resolutions in SKCM single-cell data. Heatmaps showing Spearman rank correlations of sample rankings between resolution pairs. QeITH demonstrated high stability across resolutions, while ROGUE-based heterogeneity scores showed substantially lower consistency. (D) Comparison of QeITH and ROGUE in 2 independent single-cell datasets. Boxplots with jittered points show median, IQR, and individual sample values. Left: Kidney cancer dataset (Young et al.). QeITH detected significant differences between tumor and normal tissues, while ROGUE showed no significant difference. Right: Lung cancer dataset (Maynard et al.). QeITH showed trends approaching significance for smoker versus nonsmoker and responder versus nonresponder, while ROGUE showed no significant differences. The 2-tailed Mann–Whitney U test P values are shown.

Article Snippet: We downloaded 12 scRNA-seq datasets from National Center for Biotechnology Information (NCBI) Gene Expression Omnibus (GEO) ( https://www.ncbi.nlm.nih.gov/geo/ ).

Techniques: Single Cell, Derivative Assay, RNA Sequencing, Comparison, MANN-WHITNEY

Journal: Frontiers in Immunology

Article Title: Identification of mitochondria-related biomarkers in liver fibrosis via interpretable machine learning and WGCNA: transcriptomic analysis and In Vivo validation

doi: 10.3389/fimmu.2026.1705706

Figure Lengend Snippet: Single-cell transcriptomic analysis of liver fibrosis. (A) Quality control metrics before cell filtering, including the distribution of gene counts (nFeature_RNA), UMI counts (nCount_RNA), and the percentages of mitochondrial and hemoglobin genes across samples. (B) Cell clustering of liver fibrosis samples. (C) Cell-type annotation of single-cell RNA-seq data. (D) Cell cycle analysis of single-cell transcriptomic data. (E) Proportional changes of different cell types between normal and fibrotic groups. (F) Expression distribution of Acot9, Aldh1b1, and Pck2 across different cell types.

Article Snippet: Single-cell RNA sequencing (scRNA-seq) datasets were obtained from GSE145086 and GSE233084 , both generated using the 10X Genomics platform ( , ).

Techniques: Single Cell, Control, RNA Sequencing, Cell Cycle Assay, Expressing

Journal: Nucleic Acids Research

Article Title: CSsingle: a unified tool for robust decomposition of bulk and spatial transcriptomic data across diverse single-cell references

doi: 10.1093/nar/gkag410

Figure Lengend Snippet: Schematic representation of the CSsingle workflow and performance validation. ( A ) CSsingle decomposes spatial and bulk transcriptomic data into a set of predefined cell types using the scRNA-seq or flow sorting reference. The main workflow is summarized in steps 1–5, each marked by a numbered circle. (B, C) Application of CSsingle to the deconvolution of bulk mixtures of HEK and Jurkat cells (dataset from Fig. ). ( B ) The plot illustrates the estimated cell type proportions by CSsingle compared to the actual cell type proportions, with shapes for cell types (circles: HEK; diamonds: Jurkat) and colors distinguishing samples. ( C ) Boxplot depicting mean absolute deviation (mAD) between estimated and actual cell type proportions, with colors differentiating benchmark methods. The box encompasses quartiles of mAD, and whiskers span 1.5× the interquartile range. CSsingle–ERCC: cell–size corrected via ERCC spike–ins; CSsingle: no size correction. MuSiC*: cell_size parameter estimated using ERCC spike–ins; MuSiC: cell_size estimated from data (default).

Article Snippet: To build the signature matrix, we used an independent scRNA-seq dataset of human colorectal cancer, generated using 10x Genomics [ ].

Techniques: Biomarker Discovery

Journal: Nucleic Acids Research

Article Title: CSsingle: a unified tool for robust decomposition of bulk and spatial transcriptomic data across diverse single-cell references

doi: 10.1093/nar/gkag410

Figure Lengend Snippet: CSsingle improves cross-source deconvolution. ( A ) Jitter plots displaying true and estimated cell type proportions in pancreatic islet. Each color represents a benchmarked method. Healthy subjects are denoted as dots while T2D subjects are denoted as triangles. ( B ) Decomposition benchmark of human PBMC using scRNA-seq reference data derived from six distinct scRNA-seq methods (10x Chromium v2, 10x Chromium v3, CEL-seq2, Drop-seq, inDrops, and Seq-Well).

Article Snippet: To build the signature matrix, we used an independent scRNA-seq dataset of human colorectal cancer, generated using 10x Genomics [ ].

Techniques: Derivative Assay