PLoS Computational Biology最新文献

The advent of temporal single-cell RNA sequencing (scRNA-seq) data has enabled in-depth investigation of dynamic processes in heterogeneous multicellular systems. Despite remarkable advancements in computational methods for modeling cellular dynamics, integrating cell-cell interactions (CCIs) into these models remains a major challenge. This is particularly true when dealing with high-dimensional gene expression profiles from large populations of interacting cells, where the intricate interplay between cells can be obscured by data complexity. To address this, we present scIMF, a single-cell deep-generative Interacting Mean Field model that learns collective multicellular dynamics. Leveraging the McKean-Vlasov stochastic differential equation framework, scIMF provides a mathematical foundation for describing interacting multicellular systems, where each cell's evolution depends on the population's empirical distribution. By incorporating a cell-wise attention mechanism, the model efficiently captures nonlocal and asymmetric CCIs, enabling realistic reconstruction of complex intercellular relationships in high-dimensional spaces. Benchmarking across diverse temporal scRNA-seq datasets demonstrates that scIMF outperforms state-of-the-art methods in reconstructing gene expression at unobserved time points and in inferring cellular velocities. Furthermore, scIMF uncovers biologically interpretable, non-reciprocal interaction patterns of cells, providing a principled framework for studying complex, particularly non-equilibrium biological systems.

Recent advances in digital pathology have enabled comprehensive analyses of Whole-Slide Images (WSIs) from tissue samples, leveraging high-resolution microscopy and computational capabilities. Despite this progress, available tools for automatic cell type identification perform poorly on skin tissue, e.g. in the classification of non-melanoma tumor cells. This is due to a paucity of labeled training data sets and high morphological similarities between tumor and non-tumor epithelial cells in the skin. Here, we propose Histo-Miner, a deep learning-based pipeline designed for the analysis of skin WSIs. To this end we generated two new datasets using WSIs of cutaneous Squamous Cell Carcinoma (cSCC) samples, a frequent non-melanoma skin cancer, by annotating 47,392 cell nuclei across 5 cell types in 21 WSIs and segmenting tumor regions in 144 WSIs. Histo-Miner employs convolutional neural networks and vision transformers for nucleus segmentation and classification, as well as tumor region segmentation. Performance of trained models positively compares to state of the art with multi-class Panoptic Quality (mPQ) of 0.569 for nucleus segmentation, macro-averaged F1 of 0.832 for nucleus classification and mean Intersection over Union (mIoU) of 0.907 for tumor region segmentation. From these output, the pipeline can generate a compact feature vector summarizing tissue morphology and cellular interactions, which can be used for various downstream tasks. As an exemplary use-case, we deploy Histo-Miner to predict cSCC patient response to immunotherapy based on pre-treatment WSIs from 45 patients. Histo-Miner predicts patient response with mean area under ROC curve of 0.755 ± 0.091 over cross-validation, and identifies percentages of lymphocytes, the granulocyte to lymphocyte ratio in tumor vicinity and the distances between granulocytes and plasma cells in tumors as predictive features for therapy response. This highlights the applicability of Histo-Miner to clinically relevant scenarios, providing direct interpretation of the classification and insights into the underlying biology. Importantly, Histo-Miner is designed to allow for its use on other cancer types and on other training datasets. Our tool and datasets are available through our github repository: https://github.com/bozeklab/histo-miner.

Protein molecules must efficiently locate specific DNA sequences within the densely packed chromatin of the cell nucleus. We investigate how the spatial organisation of chromatin, specifically its organisation into Topologically Associating Domains (TADs), fundamentally affects this search process. Using exact analytical theory and simulations of different models of chromatin, we show that target search within compact, highly connected chromatin domains can leverage intersegmental jumps to significantly decrease search times. Further, we establish that there exists an optimal degree of polymer compaction that minimizes the search time for proteins to find their targets. For highly folded domains, our results suggest that rather than bulk diffusion, intersegmental transfers - jumping between chromatin segments that are close together in space - drive the optimal search process. Remarkably, when we analyse 8,355 TAD structures across the human genome, we find that their natural connectivity matches with the theoretical optimum predicted by our model. The structural organisation within TADs significantly reduces protein search times far beyond what is achievable through classical facilitated diffusion. In essence, our work suggests that packaging of chromatin inside the nucleus has implications beyond spatial organisation, and is also intricately linked to the dynamics of proteins inside the nuclear environment.

Multiple sclerosis (MS) is a neurodegenerative disease in which misdirected, persistent activity of the immune system degrades the protective myelin sheaths of nerve axons. Historically, treatment of MS has relied on disease-modifying therapies that involve immunosuppression, such as targeting of the blood-brain barrier (BBB) to restrict lymphocyte movement. New therapeutic ideas in the development pipeline are instead designed to promote populations of myelin producing cells, oligodendrocytes, by exploiting their innate resilience to the stressors of MS or restoring their numbers. Given the significant advancements made in immunological disease understanding due to mathematical and computational modelling, we sought to develop a platform to (1) interrogate our understanding of the neuroimmunological mechanisms driving MS development and (2) examine the impact of different therapeutic strategies. To this end we propose a novel, open-source, agent-based model of lesion development in the CNS. Our model includes crucial populations of T cells, perivascular macrophages, and oligodendrocytes. We examine the sensitivity of the model to key parameters related to disease targets and conclude that lesion stabilisation can be achieved when targeting the integrated stress response of oligodendrocytes. Most significantly, complete prevention of lesion formation is observed when a combination of approved BBB-permeability targeting therapies and integrated-stress response targeting therapies is administered, suggesting the potential to strike a balance between a patient's immune inflammation and their reparative capacity. Given that there are many open questions surrounding the etiology and treatment of MS, we hope that this malleable platform serves as a tool for experimentalists and modellers to test and generate further hypotheses regarding this disease.

Collagen damage in articular cartilage plays a key role in post-traumatic osteoarthritis, but the underlying mechanobiological pathways leading to collagen fibril degeneration after injury remain incompletely understood. We hypothesized that mechanical injurious loading induces localized cellular damage in cartilage, which in turn triggers the release of collagen-degrading matrix metalloproteinases (MMPs) and depth-wise collagen loss. To investigate this, we developed a computational mechano-signaling model for injured bovine cartilage, in which injury-induced cell damage is caused by excessive localized shear strains, leading to downstream MMP release, and spatially heterogeneous collagen degradation. The model predictions were compared to ex vivo cartilage explant experiments over 12 days post-injury. By day 12, the simulated bulk and depth-wise collagen loss aligned with our experimental findings quantified via Fourier-transform infrared microspectroscopy imaging (~30% average loss in the model vs. ~ 35% in the experiment). The results suggest that injury-induced cell damage and the downstream MMP activity can partly explain the depth-wise collagen content loss observed in the early days after cartilage injury. Ultimately, combining the current mechanistic approach with joint-level computational models could enhance the prediction of the onset and progression of cartilage degeneration following joint trauma.

Theories of predictive processing propose that sensory systems constantly predict incoming signals, based on spatial and temporal context. However, evidence for prediction in sensory cortex largely comes from artificial experiments using simple, highly predictable stimuli, that arguably encourage prediction. Here, we test for sensory prediction during natural scene perception. Specifically, we use deep generative modelling to quantify the spatial predictability of receptive field (RF) patches in natural images, and compared those predictability estimates to brain responses in the mouse visual cortex-while rigorously accounting for established tuning to a rich set of low-level image features and their local statistical context-in a large scale survey of high-density recordings from the Allen Institute Brain Observatory. This revealed four insights. First, cortical responses across the mouse visual system are shaped by sensory predictability, with more predictable image patches evoking weaker responses. Secondly, visual cortical neurons are primarily sensitive to the predictability of higher-level image features, even in neurons in the primary visual areas that are preferentially tuned to low-level visual features. Third, unpredictability sensitivity is stronger in the superficial layers of primary visual cortex, in line with predictive coding models. Finally, these spatial prediction effects are independent of recent experience, suggesting that they rely on long-term priors about the structure of the visual world. Together, these results suggest visual cortex might predominantly predict sensory information at higher levels of abstraction-a pattern bearing striking similarities to recent, successful techniques from artificial intelligence for predictive self-supervised learning.

Eukaryotic genomes are organized within nuclei in three-dimensional space, forming structures such as loops, topologically associating domains (TADs), and chromosome territories. This 3D architecture impacts gene regulation and development, stress responses, and disease. However, current methods to infer these 3D structures from genomic data have multiple drawbacks, including varying outcomes depending on the resolution of the analysis and sequencing depth, qualitative outputs that limit statistical comparisons, and insufficient insight into structure frequency within samples. These challenges hinder rigorous comparisons of 3D properties across genomes, conditions, or species. To overcome these issues, we developed WaveTAD, a wavelet transform-based method that provides a resolution-free, probabilistic, and hierarchical description of 3D organization. WaveTAD generates TAD strengths, capturing the variable frequency of intrachromosomal interactions within samples, and shows increased accuracy and sensitivity over existing methods. We applied WaveTAD to multiple datasets from Drosophila, mouse, and humans to illustrate new biological insights that our more sensitive and quantitative approach provides, such as the widespread presence of embryonic 3D organization before zygotic genome activation, the effect of multiple CTCF units on the stability of loops and TADs, and the association between gene expression and TAD structures in COVID-19 patients or sex-specific transcription in Drosophila.

Patient-derived cells (PDC) mouse xenografts are increasingly important tools in glioblastoma (GBM) research, essential to investigate case-specific growth patterns and treatment responses. Despite the central role of xenograft models in the field, few good simulation models are available to probe the dynamics of tumor growth and to support therapy design. We therefore propose a new framework for the patient-specific simulation of GBM in the mouse brain. Unlike existing methods, our simulations leverage a high-resolution map of the mouse brain anatomy to yield patient-specific results that are in good agreement with experimental observations. To facilitate the fitting of our model to histological data, we use Approximate Bayesian Computation. Because our model uses few parameters, reflecting growth, invasion and niche dependencies, it is well suited for case comparisons and for probing treatment effects. We demonstrate how our model can be used to simulate different treatment by perturbing the different model parameters. We expect in silico replicates of mouse xenograft tumors can improve the assessment of therapeutic outcomes and boost the statistical power of preclinical GBM studies.

Drugs play a central role in limiting bacterial population spread, yet laboratory studies typically assume well-mixed environments when assessing microbial drug responses. In contrast, bacteria in the human body often occupy spatially structured habitats where drug concentrations vary. Understanding how this heterogeneity shapes growth and decline is therefore essential for controlling infections and mitigating resistance evolution. Here, we developed a minimal robot-automated system to study how spatial drug heterogeneity affects short-term population dynamics in E. faecalis, a Gram-positive opportunistic pathogen. This system was combined with a theoretical framework to interpret and explain the observed outcomes. We first recapitulated the classic critical-patch-size model result: in a spatially homogeneous environment, a population persists in a finite domain only when growth outpaces diffusive losses at the boundaries. In heterogeneous environments, we found certain conditions that population persistence can depend critically on the spatial arrangement of the drug, even when its total amount is fixed. Using theoretical and experimental approaches, we identified the arrangements that produce the strongest growth and the fastest decline, revealing the range of possible outcomes under drug heterogeneity. We further tested this framework in more complex environments, including ring-shaped communities, and observed consistent arrangement-dependent behavior. Overall, our results extend the classical growth-condition framework to general heterogeneous environments and demonstrate that spatial drug arrangement - not only total dose - can strongly influence bacterial population dynamics. These findings highlight the importance of spatially structured dosing strategies and motivate further theoretical and experimental investigation.

[This corrects the article DOI: 10.1371/journal.pcbi.1013452.].