Journal: Advanced Science

Article Title: scPER: A Rigorous Computational Approach to Determine Cellular Subtypes in Tumors Aligned With Cancer Phenotypes From Total RNA Sequencing

doi: 10.1002/advs.202514502

Figure Lengend Snippet: Overview of scPER and the function of each stage. Stage 1. Reference integration and preprocessing. scRNA‐seq datasets from multiple studies or tissues are integrated to form a reference matrix R (rows = genes, columns = cells). After QC and feature selection, denoising and imputation are applied to stabilize gene signals across datasets. Output: preprocessed single‐cell expression R with cell‐type labels and confounder labels (study/platform/tissue). Stage 2. Confounder‐aware representation learning. An adversarial autoencoder learns a low‐dimensional latent space Z from R . The encoder E maps expression x to z ; the decoder D reconstructs x from z . An adversary V tries to predict designated confounders c (e.g., study/platform/tissue) from z . Training minimizes reconstruction while implicitly maximizing adversary loss with weight λ (i.e., L rec − λ L adv ), yielding Z that preserves biology but reduces nuisance variation. Output: cell embeddings Z and a trained encoder E for embedding new data. Stage 3. Bulk embedding and proportion estimation. Bulk RNA‐seq samples T are normalized like the single‐cell data and passed through the trained encoder E to obtain bulk embeddings in the same latent space. An XGBoost regressor is trained on latent features from single‐cell references to predict cell‐type compositions P for each bulk sample (columns = samples, rows = cell types). Output: estimated cell‐type proportions P across all bulk samples. Stage 4. Downstream biology and clinical association. The estimated compositions P enable downstream analyses: (i) identify critical genes modulating the cell proportions, (ii) test clinical endpoints such as survival or response to immunotherapy, and (iii) identify phenotype‐associated changes in cell types or subclusters. R = reference single‐cell matrix; T = bulk expression; Z = scPER latent embeddings; P = estimated cell‐type proportions; λ = adversarial weight; c = confounder label. Arrows indicate data flow from reference construction to embedding, deconvolution, and phenotype analyses.

Article Snippet: In this study, the PBMC scRNA‐seq dataset in Figure was downloaded from the 10X Genomics data download page: 8k PBMCs from a healthy donor ( https://www.10xgenomics.com/resources/datasets/8‐k‐pbm‐cs‐from‐a‐healthy‐donor‐2‐standard‐2‐1‐0 ).

Techniques: Selection, Single Cell, Expressing, RNA Sequencing

Journal: Advanced Science

Article Title: scPER: A Rigorous Computational Approach to Determine Cellular Subtypes in Tumors Aligned With Cancer Phenotypes From Total RNA Sequencing

doi: 10.1002/advs.202514502

Figure Lengend Snippet: Cross‑tissue deconvolution performance comparison. (A) Workflow and latent embedding of the cross‑tissue reference panel. Left: UMAP of single‑cell transcriptomes from melanoma, ovarian cancer ascites, and PBMC sources before batch correction. Top subpanel: cells colored by tissue of origin; bottom subpanel: cells colored by annotated cell type. Middle: heatmap of scPER's 100 latent representations after adversarial batch correction, with rows as embedding features and columns as six cell types. (B) Pearson correlation coefficients between true and predicted cell‑type fractions for each deconvolution tool, summarizing overall accuracy across the 16 simulated prostate cancer bulk samples. (C‐F) True versus predicted proportion scatter plots for monocytes/macrophages using (C) scPER, (D) CIBERSORTx, (E) BayesPrism, and (F) Scaden. Points represent individual samples; the x‑axis shows ground‑truth fractions (from summed single‑cell counts), and the y‑axis shows model predictions.

Article Snippet: In this study, the PBMC scRNA‐seq dataset in Figure was downloaded from the 10X Genomics data download page: 8k PBMCs from a healthy donor ( https://www.10xgenomics.com/resources/datasets/8‐k‐pbm‐cs‐from‐a‐healthy‐donor‐2‐standard‐2‐1‐0 ).

Techniques: Comparison

Journal: Advanced Science

Article Title: scPER: A Rigorous Computational Approach to Determine Cellular Subtypes in Tumors Aligned With Cancer Phenotypes From Total RNA Sequencing

doi: 10.1002/advs.202514502

Figure Lengend Snippet: scPER‐derived cell‐type proportions predict melanoma immunotherapy outcomes. (A‐B) UMAP projections of scRNA‐seq from two melanoma datasets before scPER integration. Colored by annotated cell types. Red rectangles highlight cell populations unique to one study.(C) Boxplots comparing estimated proportions of each cell type between responders (R; n=14) and non‐responders (NR; n = 13) to immunotherapy. P‐values were calculated using the Wilcoxon rank‐sum test. (D) Receiver operating characteristic (ROC) curves for XGBoost classifier predicting response status: one using scPER‐estimated cell‐type proportion features alone and another using tumor mutation burden (mutation load) alone. Area under the ROC curve (AUROC) values are indicated for each model. (E) Kaplan‐Meier survival analysis of LTB expression in the TCGA‐SKCM cohort. Patients were stratified into high‐ and low‐ LTB expression groups based on the median; the log‐rank test p‐value is shown.

Article Snippet: In this study, the PBMC scRNA‐seq dataset in Figure was downloaded from the 10X Genomics data download page: 8k PBMCs from a healthy donor ( https://www.10xgenomics.com/resources/datasets/8‐k‐pbm‐cs‐from‐a‐healthy‐donor‐2‐standard‐2‐1‐0 ).

Techniques: Derivative Assay, Mutagenesis, Expressing

Journal: Advanced Science

Article Title: scPER: A Rigorous Computational Approach to Determine Cellular Subtypes in Tumors Aligned With Cancer Phenotypes From Total RNA Sequencing

doi: 10.1002/advs.202514502

Figure Lengend Snippet: Benchmarking deconvolution accuracy on PBMC single‐cell and bulk profiles.(A) t‐Distributed Stochastic Neighbor Embedding (t‐SNE) of the 100 latent embeddings learned from scRNA‐seq data, with points colored by the six annotated immune cell types. (B) Pearson correlation coefficients ( r ) between true and predicted cell‐type fractions across all bulk samples, shown for each deconvolution tool as a violin plot summarizing the distribution of r values. (C) True versus predicted proportion scatter plots for each cell type and method. Each point represents one bulk sample (n = 164); the x‐axis shows the ground‐truth fraction (flow cytometry), and the y‐axis shows the fraction predicted by the deconvolution tool.

Article Snippet: In this study, the PBMC scRNA‐seq dataset in Figure was downloaded from the 10X Genomics data download page: 8k PBMCs from a healthy donor ( https://www.10xgenomics.com/resources/datasets/8‐k‐pbm‐cs‐from‐a‐healthy‐donor‐2‐standard‐2‐1‐0 ).

Techniques: Single Cell, Flow Cytometry

Journal: Genomics, Proteomics & Bioinformatics

Article Title: scEMAIL: Universal and Source-free Annotation Method for scRNA-seq Data with Novel Cell-type Perception

doi: 10.1016/j.gpb.2022.12.008

Figure Lengend Snippet: Comparison of model performance for single-omics and multi-omics experiments on paired data from 10x Genomics 10 k PBMC dataset

Article Snippet: In order to verify whether other omics data can assist scRNA-seq data in cell-type annotation and novel cell-type identification, we applied paired data from 10x 10 k peripheral blood mononuclear cell (PBMC) dataset, which was downloaded from 10x Genomics website ( https://www.10xgenomics.com/resources/datasets/10-k-pbm-cs-from-a-healthy-donor-gene-expression-and-cell-surface-protein-3-standard-3-0-0 ), which provided additional phenotypic information by integrating measurements of cellular proteins and transcriptomes of single cells.

Techniques: Comparison, Biomarker Discovery

Journal: Genomics, Proteomics & Bioinformatics

Article Title: scEMAIL: Universal and Source-free Annotation Method for scRNA-seq Data with Novel Cell-type Perception

doi: 10.1016/j.gpb.2022.12.008

Figure Lengend Snippet: M apping between real cell types and predicted cell types on single-omics and multi-omics experiments via Sankey plots These two tasks are applied on the paired data from 10x Genomics 10 k PBMC dataset downloaded from 10x Genomics website ( https://www.10xgenomics.com/resources/datasets/10-k-pbm-cs-from-a-healthy-donor-gene-expression-and-cell-surface-protein-3-standard-3–0-0 ) with cell types “CD14 + monocytes” and “CD4 + T cells” set as novel cell types, respectively. For experiments with novel cell type “CD14 + monocytes”, we show the annotation results of source model only on single-omics ( A ) and multi-omics ( B ), as well as the results of scEMAIL on single-omics ( C ) and multi-omics ( D ) data. For experiments with novel cell type “CD4 + T cells”, the corresponding results of source model only on single-omics ( E ) and multi-omics ( F ), as well as the results of scEMAIL on single-omics ( G ) and multi-omics ( H ) data, are also exhibited. NK, natural killer; PBMC, peripheral blood mononuclear cell.

Article Snippet: In order to verify whether other omics data can assist scRNA-seq data in cell-type annotation and novel cell-type identification, we applied paired data from 10x 10 k peripheral blood mononuclear cell (PBMC) dataset, which was downloaded from 10x Genomics website ( https://www.10xgenomics.com/resources/datasets/10-k-pbm-cs-from-a-healthy-donor-gene-expression-and-cell-surface-protein-3-standard-3-0-0 ), which provided additional phenotypic information by integrating measurements of cellular proteins and transcriptomes of single cells.

Techniques: Biomarker Discovery, Gene Expression