Immune biomarkers from bulk RNA-sequencing show inconsistent performance across cancer cohorts. We hypothesized that we can anticipate failure by characterizing transcriptomic heterogeneity within cellular sources and its interactions with the tumor microenvironment. We developed two metrics: Source Fidelity Index (SFI), quantifying T/NK-specific expression from 2.2 million single cells across 18 datasets, and Biomarker Fragility Index (BFI), quantifying composition-sensitivity through 3D simulation across 82 synthetic compositions per gene. We validated a six-factor Operational Risk Stratification framework in 754 samples from 8 immune checkpoint blockade cohorts and evaluated BFI-class predictions of clinical response in 10 independent external cohorts spanning 5 cancer types (989 patients). Of 101 immune genes, only 35 (34.7
DNA double-strand breaks (DSBs), visualized as ionizing radiation-induced repair foci (IRIFs), represent a severe form of DNA damage and are central to radiation biology, cancer research, and space health. Quantifying IRIF induction and repair is critical for biodosimetry but is traditionally performed using approaches sensitive to image variability. This study developed a robust and scalable method for accurate IRIF detection and quantification using deep learning. A modified U-Net architecture was employed for semantic segmentation of IRIFs in fluorescence microscopy images. The model was trained on 95,000 synthetically generated image-mask pairs designed to replicate signal, noise, and morphological characteristics of real microscopy data. Model performance was evaluated both on synthetic data and on a real-world dataset of 345 manually annotated mouse fibroblast images from the NASA GeneLab OSD-366 experiment, encompassing multiple radiation types and doses. Segmentation and foci detection performance were assessed using Dice coefficient, Intersection over Union (IoU), precision and recall. On synthetic test data, the model achieved a mean Dice coefficient of 0.995 and IoU of 0.993. On real-world data, Dice coefficients ranged from 0.72 to 0.88 on average across radiation categories. Foci detection demonstrated high sensitivity (0.95–0.98) and precision (0.82–0.97), with overall recall of 0.97 and precision of 0.89, indicating robust detection with limited false negatives. This work demonstrates that training on synthetic data enables reliable generalization to real radiation biology images. The proposed approach improves reproducibility and robustness,and may support high-throughput biodosimetry, with potential implications for radiation protection, personalized medicine, and astronaut health assessment, pending further validation across additional time points, cell types, and data sources.
Single-cell RNA sequencing (scRNA-seq) enables the reconstruction of cell state transitions from static snapshots through pseudotime trajectory analysis. Although existing methods can infer tree-structured lineages to characterize developmental hierarchies, they often generate multifurcating or high-degree branching patterns that are difficult to interpret biologically. To address this limitation, we propose an integer programming based degree-restricted minimum spanning tree (MST) algorithm for binary-structured cell trajectory inference, which aligns more closely with Waddington’s epigenetic landscape for cell fate decisions. To facilitate deployment, our method is implemented as a highly flexible plug-in module that can be seamlessly integrated into existing graph- and tree-based trajectory inference pipelines. We demonstrate its integration into Slingshot for pseudotime inference and evaluate the consistency of the resulting constrained topology against a CoSpar-derived reference hierarchy. Results on bone marrow mononuclear cell, human fetal immune cell, and mouse hematopoiesis datasets illustrate the interpretability of the constrained trajectories and provide a descriptive comparison with unconstrained MST baselines. The source code is available at https://github.com/iwakitakuma33/degree_restricted_slingshot.
The ability to capture cellular dynamics at high spatial and molecular resolution is crucial for understanding complex biological processes, including development, tissue remodeling, and disease progression. Single-cell RNA sequencing (scRNA-seq) provides detailed transcriptional profiles of individual cells, while spatial transcriptomics (ST) preserves their spatial organization within tissues. However, integrating these complementary modalities across multiple time points to reconstruct temporal changes in spatial cell-type organization and cell-state composition remains a major challenge. Here, we present a computational framework that integrates time-series scRNA-seq and spatial transcriptomics data to reconstruct spatiotemporal cell-type redistribution and inferred cell-state dynamics. The framework combines reference-based spatial deconvolution with single-cell spatial mapping to generate temporally resolved representations of cellular organization. Temporal changes in spatial cell-type distributions are quantified using complementary abundance-weighted spatial descriptors, whereas inferred cell-state dynamics are characterized through changes in dominant cell-type assignments across consecutive time points. We further evaluated the spatial abundance component of the framework using simulation-based quantitative benchmarking against alternative spatial deconvolution and mapping approaches, using known simulated cell-type compositions as ground truth. We demonstrate the utility and generalizability of the framework using two biologically distinct systems: human glioblastoma organoids (GBOs) and embryonic chicken heart development. Across both datasets, the framework captures temporal patterns of spatial cell-type organization and inferred cell-state composition that are not accessible through static analyses alone. This work provides a broadly applicable computational framework for investigating temporal tissue organization through the integration of single-cell and spatial transcriptomic data, thereby enabling systematic characterization of dynamic cellular behaviors across developmental and disease-related processes. Quantitative benchmarking supports the ability of the framework to recover simulated spatial cell-type abundance patterns, while the inferred redistribution and cell-state dynamics should be interpreted as descriptive computational measures rather than direct evidence of physical cell migration or lineage-validated state transitions.
Ageing-associated plasma proteomic signatures can predict chronological age, but translating predictive biomarkers into mechanistically interpretable and experimentally testable target hypotheses remains challenging. We developed Protein Expression Net (PEN), a six-stage computational framework for linking plasma proteomic ageing signatures to candidate mechanistic target hypotheses. PEN integrates six sequential analytical stages: (1) plasma proteome input, (2) proteomic age prediction using a multilayer perceptron (MLP), (3) diagnostic protein identification and functional module characterization using Integrated Gradients, (4) graph-based target discovery using a protein–protein interaction network with graph convolutional propagation and diffusion, (5) discovery-oriented reranking integrating multi-component biological evidence, and (6) Geneformer-based virtual perturbation as an orthogonal in silico support stage. Applied to 44,179 UK Biobank participants profiled for approximately 2,920 plasma proteins, the MLP achieved accurate age prediction (test R² = 0.8723, Pearson r = 0.934, MAE = 2.30 years). Graph-based prioritization identified candidate mechanistic target hypotheses including FBN1, FBLN5, EDA, RLN3, LHCGR, COL14A1, HLA-E, and TSPAN4. Geneformer-based virtual perturbation in an independent ageing human skin single-cell transcriptomic dataset (GSE130973) provided orthogonal in silico support for selected candidate hypotheses. After recalibration using an expanded null distribution of 500 random token-valid genes, FBLN5 and HLA-E showed significant positive OLD-to-YOUNG transcriptional state shifts in specific cell populations, whereas other candidates displayed weaker or context-dependent effects. PEN provides a reproducible six-stage computational framework for translating plasma proteomic ageing signatures into interpretable candidate mechanistic target hypotheses. The final Geneformer perturbation stage provides orthogonal in silico functional support, strengthening biological plausibility without constituting experimental validation. The framework is generalizable to other biomarker-driven target discovery problems.
