Biology subjects

Sakumura, Y.

Publications and source records attributed to Sakumura, Y..

10 recordsLinked to original sources

bioRxiv2026

Expanding gene regulatory networks from transcriptome data through graphical modeling with heterogeneous priors

Gene regulatory network inference is widely used to reconstruct large-scale networks and identify functional genes from transcriptome data. Meanwhile, in many biological fields, core regulatory genes have been extensively studied, leading to the establishment of small-scale gene regulatory networks, and novel genes connected to these networks remain to be identified. However, methods for expanding existing gene networks by identifying novel regulatory interactions, rather than reconstructing the entire network, are not well established. Here, we propose a method for gene network expansion that incorporates known regulatory relationships and evaluates each candidate gene individually to infer its regulatory connections to the existing network. Using simulated datasets from the DREAM4 benchmark and the PRECISE-1K experimental dataset, our method outperformed conventional methods by incorporating prior knowledge. In particular, it improved the ability to distinguish true regulatory interactions from indirect associations arising from strong correlations among genes in the existing network. The method also showed strong performance for interactions involving genes with high outdegree or centrality. Furthermore, it maintained stable performance as the size of the existing network increased and was robust to noise in prior information. These results demonstrate that our method provides an effective framework for expanding existing gene regulatory networks by leveraging prior knowledge.

bioinformatics↗

bioRxiv2025

Mechanical polarity links adhesion-tuned protrusions to directional stability in glioblastoma cell migration

Glioblastoma invasion critically limits therapeutic outcomes, and understanding the physical principles that govern cell motility is essential for developing effective therapies. Here, to clarify the mechanical links between cell adhesion, protrusions, and migration, we analyzed glioblastoma-derived cells migrating on fibronectin- and laminin-coated extracellular matrix (ECM) substrates using time-lapse imaging and mathematical modeling. We quantified cell motility on each ECM and constructed a coarse-grained biophysical model that incorporates catch- and slip-bond kinetics. We treated the ECM as an external boundary condition that modulates adhesion dynamics, distinct from the intrinsic mechanical parameters of the cell. The model reproduced ECM-dependent differences in cell motility. The results suggest that adhesions and protrusion elongation on fibronectin are less stable than on laminin, and opposing forces between protrusions reduce net displacement more strongly on fibronectin than on laminin, leading to less coordinated tensile forces. Based on these results, we establish mechanical polarity, defined as an imbalance of protrusion- and adhesion-mediated forces that drive directional migration, as a quantifiable physical principle of cell protrusion and migration. This principle likely extends beyond glioblastoma biology, providing a generalizable mechanism that links adhesion dynamics to migration stability and offering a physical basis for strategies to suppress invasive cell behavior.

systems biology↗

bioRxiv2025

Non-invasive estrous cycle classification in mice using convolutional neural networks

Accurately identifying the estrous cycle phases in laboratory mice is essential for neurological research, reproductive studies, and breeding programs. Here, we introduce a non-invasive approach using Convolutional Neural Networks (CNNs) to classify estrous stages from external genital images, with cytological smears serving as the reference standard for labeling. Images were curated, preprocessed, and augmented to ensure consistency and generalization. Seven pre-trained CNNs and a lightweight custom architecture, Repro Cycle Net (RCN), were evaluated. All models achieved accuracies above 75%, with RCN exceeding 83% and showing the lowest test loss, highlighting its efficiency despite a simple four-layer design. Saliency map analyses revealed that classification relied on perivaginal features, while irrelevant regions such as the fur area were largely ignored. Importantly, binary classification of estrus versus non-estrus directly informs mating feasibility, underscoring the immediate utility of this method for reproductive studies and colony management. This work demonstrates that combining deep learning with external genital observation enables efficient and reproducible estrous monitoring, supporting both experimental reliability and animal welfare.

bioinformatics↗

bioRxiv2025

Cell Morphology and Biophysical Mechanisms-Informed Traction Force Microscopy Using Machine Learning

AO_SCPLOWBSTRACTC_SCPLOWQuantitative evaluation of cell motility is essential for understanding biological functions. Traction force microscopy (TFM) is a method for quantifying cellular traction forces. By performing inverse mathematical analysis of substrate deformations induced by cellular forces, it is possible to estimate the underlying traction forces. Among various approaches for detecting substrate deformations, the most widely used is to track the displacement of fluorescent beads randomly embedded in the substrate. However, it is well known that the accuracy of force estimation deteriorates when the observation density is low. Furthermore, standardized datasets have not been established, as validation settings--such as random force distributions on the substrate, cell size, and bead density--vary across studies. To enable quantitative evaluation, we constructed a dataset based on the biophysical mechanism by which cellular traction forces are transmitted to the substrate through stress fibers, clutch proteins, and focal adhesions. In addition, we proposed a novel machine learning model that incorporates cell shape information obtained from observations, and we designed a loss function that accounts for both force magnitude and direction. As a result, our method enables highly accurate force estimation even under sparse observation conditions.

biophysics↗

bioRxiv2024

Stepwise Bayesian Machine Learning Uncovers a Novel Gene Regulatory Network Component in Neural Tube Development

Recent advancements in machine learning-based data processing techniques have facilitated the inference of gene regulatory interactions and the identification of key genes from multidimensional gene expression data. In this study, we applied a stepwise Bayesian framework to uncover a novel regulatory component involved in differentiation of specific neural and neuronal cells. We treated naive neural precursor cells with Sonic Hedgehog (Shh) at various concentrations and time points, generating comprehensive whole-genome sequencing data that captured dynamic gene expression profiles during differentiation. The genes were categorized into 224 subsets based on their expression profiles, and the relationships between these subsets were extrapolated. To accurately predict gene regulation among subsets, known networks were used as a core model and subsets to be added were tested stepwise. This approach led to the identification of a novel component involved in neural tube patterning within gene regulatory networks (GRNs), which was experimentally validated. Our study highlights the effectiveness of in silico modeling for extrapolating GRNs during neural development.

systems biology↗

bioRxiv2024

Designing a hybrid in silico/in-cell controller robust to process-model mismatch associated with dynamically regulated enzyme overexpression

Discrepancy between model predictions and actual processes, known as process-model mismatch (PMM), remains a serious challenge in bioprocess optimization. Previously, we proposed the hybrid in silico/in-cell controller (HISICC) concept combining model-based optimization with cell-based feedback to address the PMM problem. Herein, this approach was advanced to regulate intracellular concentrations of rate-limiting enzymes. Advanced HISICC was demonstrated using an engineered Escherichia coli strain for fatty acid production (FA3). This strain harbors an in-cell feedback controller, which decelerates acetyl-CoA carboxylase (ACC) overexpression in response to sensing the concentration of malonyl-CoA formed by this enzyme. A mathematical model for FA3 was constructed and validated using experimental data. Simulations assuming various PMM revealed that the HISICC using FA3 effectively mitigates toxicity from excessive ACC by robustly adapting braking its overexpression, minimizing yield loss. This study confirmed HISICC as a viable strategy for enhancing bioprocess efficiency, especially in balancing the bottleneck enzyme levels.

synthetic biology↗

bioRxiv2024

Cell morphogenesis via shape-guided self-propelled treadmilling actin waves

Cells frequently undergo spontaneous morphogenesis, yet the underlying mechanisms remain incompletely understood. While actin filaments are central to cell morphogenesis and are typically regulated by biochemical signaling, cells can form protrusions even without clear external cues, suggesting the existence of intrinsic physical mechanisms. Here, we report that actin filaments undergo directional movement driven by treadmilling, an ATP-fueled polymerization-disassembly cycle intrinsic to actin. These Self-propelled Treadmilling Actin filaments (SpTAs), exhibit stochastic yet directional motion, in a manner similar to self-propelled "particles" rather than the previously reported reaction-diffusion "waves". SpTA arrival at the cell periphery drives membrane protrusion by orienting their polymerizing ends outwards. This SpTA accumulation, guided by nascent membrane curvature, further amplifies protrusion growth and expansion, driving cellular polarization for migration. Our findings establish actin filament as a novel class of active particle, providing a fundamental physical framework for understanding how molecular-scale motion leads to higher-order organization in living systems.

cell biology↗

bioRxiv2023

A hybrid in silico/in-cell controller for microbial bioprocesses with process-model mismatch

The optimization of bioprocess inputs using mathematical models is widely practiced. However, the mismatch between model prediction and the actual process [called process-model mismatch (PMM)] is problematic; when a large PMM exists, the process inputs optimized using the mathematical model in advance are no longer optimal for the actual process. In this study, we propose a hybrid control system that combines model-based optimization (in silico feedforward controller) and feedback controllers using synthetic genetic circuits integrated into cells (in-cell feedback controller) - which we named the hybrid in silico/in-cell controller (HISICC) - as a solution to this PMM issue. As a proof of concept for HISICC, we constructed a mathematical model of an engineered Escherichia coli strain for the isopropanol production process that was previously developed. This strain contains an in-cell feedback controller, and its combination with an in silico controller can be regarded as an example of HISICC. We demonstrated the robustness of HISICC against PMM by comparing the strain with another strain with no in-cell feedback controller in simulations assuming PMM of various magnitudes.

synthetic biology↗

bioRxiv2022

On-line reoptimization of mammalian fed-batch culture using a nonlinear model predictive controller

Fed-batch culture enables high productivity by maintaining low substrate concentrations in the early stage of the culture to suppress the accumulation of by-products that are harmful to cell growth. Therefore, they are widely used in the production of biopharmaceuticals by mammalian cells. However, there exists a trade-off in the design of the fed-batch process: early feeding results in the accumulation of harmful by-products, whereas late feeding results in a shortage of substrates needed for cell growth and synthesis of the desired product. To manage this trade-off and maximize the product yield, model-based optimization of the feeding trajectory has been reported in several studies. A significant drawback of this off-line optimization approach is the mismatch between the predictions made using the model and the actual process states, called the process-model mismatch (PMM). If the PMM is large, the off-line optimized feeding trajectory is no longer optimal for the actual process, resulting in lower product yields. Mammalian cell culture models typically contain dozens of unknown parameters that must be estimated prior to optimization. Sufficient parameter estimation is often unachievable owing to the nonlinear nature of these models. We believe that reoptimizing the feeding trajectory in real time using a nonlinear model predictive controller (NLMPC) is an effective solution to this PMM. Although NLMPC is a model-based feedback controller widely utilised in mammalian fed-batch culture, only a few studies have applied it to on-line reoptimization, and it remains unclear whether NLMPC with a standard kinetic model can effectively compensate for a large PMM. In this study, we demonstrated the reoptimization of the feeding trajectory with a NLMPC using two previously reported standard monoclonal antibody (mAb) production models. In both models, NLMPC successfully suppressed the reduction in mAb yield caused by the intentional introduction of PMM.

bioengineering↗

bioRxiv2021

Decoding cellular deformation from pseudo-simultaneously observed Rho GTPase activities

The inability to simultaneously observe all of the important Rho GTPases (Cdc42, Rac1, and RhoA) has prevented us from obtaining evidence of their coordinated regulation during cell deformation. Here, we propose Motion-Triggered Average (MTA), an algorithm that converts individually observed GTPases into pseudo-simultaneous observations. Using the time series obtained by MTA and mathematical model, we succeeded for the first time in decoding the cell edge velocity from the three GTPase activities to provide clear numerical evidence for coordinated cell edge regulation by the three GTPases. We found that the characteristics of the obtained activities were consistent with those of previous studies, and that GTPase activities and their derivatives were involved in edge regulation. Our approach provides an effective strategy for using single-molecule observations to elucidate problems hampered by the lack of simultaneous observations.

systems biology↗