An-Chi Wei

Mitochondrial toxicity is a major concern in drug development and safety assessment. Machine learning models trained on mitochondrial membrane potential (MMP) assay data offer promising toxicity predictions, but cross-scaffold and cross-assay transferability remain uncharacterized. We evaluated classical machine learning and deep learning architectures across six molecular representations and both random and scaffold-based splitting strategies, assessing performance on an internal MMP test set and an independent external set comprising flux and glucose–galactose assay data. Across all model configurations, we observed a consistent cross-assay generalization gap, independent of model type, feature choice, or augmentation strategy. Mordred descriptors provided the most transferable predictive signal, outperforming fingerprint-based representations on the external set. SHAP analysis of the best CatBoost and Random Forest models identified autocorrelation-family descriptors as dominant predictors. C3SP3 contributed through feature interactions while ATS5i showed positive association with toxicity across assays. Motif-level and descriptor-level associations proved to be strongly assay-dependent, supporting a mechanism in which mitochondrial toxicity arises from multivariate physicochemical interactions rather than single structural alerts.

Photobiomodulation (PBM) has been shown to enhance cell growth and differentiation. This study investigates the effects of 810-nm PBM on the proliferation and cardiac differentiation of human-induced pluripotent stem cells (hiPSCs), utilizing a custom-made LED device. This device is compatible with standard cell incubators and features an LED array that ensures uniform light exposure to cells in each well of a 96-well plate. hiPSCs were irradiated with 810-nm light at intensities of 1.0, 1.5, and 2.0 mW/cm² for 15 min daily. Outcomes assessed included cell viability, mitochondrial function and morphology, and cardiomyocyte differentiation. The results show that PBM at 1.5 mW/cm² enhanced hiPSC proliferation and cardiac differentiation, yielding a higher number of spontaneously beating cardiomyocytes compared to controls. These findings suggest that 810-nm PBM may improve the efficiency of cardiac cell generation for both research and therapeutic applications.

U-Net, a fully convolutional neural network (FCNN) with U-shaped features, has demonstrated significant success in biomedical image segmentation. However, the locality of convolution operations in the U-Net limits its ability to learn long-range dependencies. Transformers, originally developed for natural language processing, have recently been adapted for image segmentation because of their global self-attention mechanisms. Inspired by the long-range feature learning capability of transformers, we propose Dense-Transformer (DenT), an architecture designed for volumetric microscopy image segmentation. DenT incorporates transformers as encoders within each convolutional layer to capture global contextual information. Additionally, dense skip connections at multiple resolutions enhance feature propagation, enabling precise localization. We evaluated DenT on mitochondrial segmentation using our confocal microscopy dataset and a public fluorescence microscope dataset from the Allen Institute for Cell Science. The experimental results demonstrate that DenT incrementally improves the segmentation of mitochondria and mitochondrial DNA substructures from transmitted light microscopy images. DenT offers a tool for visualization, measurement, and analysis of mitochondrial morphology and mitochondrial DNA in label-free microscopy.

Doxorubicin (DOX), although effective in treating cancer, has significant cardiac side effects, which limit its clinical utility. In this study, we collected time-course transcriptomics and metabolomics data from the human cardiomyocyte cell line AC16, which we analyzed along with curated public transcriptomics data on DOX-induced toxicity. We developed a multiomics analysis workflow and a computational toolbox, pipeGEM, to integrate RNA-seq data with metabolic models, enabling the simulation of DOX-induced metabolic perturbations at a sample-specific level. Our results revealed that DOX affected mitochondrial damage and mitochondria-to-nucleus retrograde signaling, potentially contributing to the observed cellular enlargement, senescence and metabolic shift. Cardiac cells that survived DOX treatment presented elevated glycolysis, increased pentose phosphate pathway activity, an altered TCA cycle, and modified glutathione and fatty acid metabolism. These findings provide a comprehensive understanding of DOX-induced toxicity and its implications for cardiac hypertrophy, suggesting potential strategies to mitigate side effects while retaining the anticancer efficacy of DOX.

An accurate and precise mechanism for measuring CaMKII activity in living cells is invaluable in the search for effective and targeted CaMKII-based therapeutics. Here, we employ our recently published CaMKII Activity Reporter (CaMKAR) biosensor in order to investigate the spatiotemporal dynamics of CaMKII activation in three different types of cells – cardiac myocytes, skeletal myocytes, and neurons. In doing so, we found a greater rate of CaMKII activation in skeletal muscle compared to cardiac muscle and also delineated CaMKAR’s ability to measure discrete CaMKII activation events in the presence of individual action potentials. By modifying the original CaMKAR sequence, we generated sensors that can be localized to subcellular compartments and thereby preferentially detect the activity of specific spatially-distributed CaMKII isoforms. Finally, we utilized the live-cell data to generate mathematical models of CaMKII activation kinetics, both as an integrated function across multiple calcium transients and as discrete on-off events following individual depolarizations. By furthering our understanding of CaMKII activity profiles across cell types and within subcellular compartments, we hope to support development of CaMKII inhibitors that are optimally precise and potent.

Animal growth is driven by the collective actions of cells, which are reciprocally influenced in real-time by the animal’s overall growth state. Whereas cell behavior and animal growth state are expected to be tightly coupled, it is not yet determined whether local cellular features at the micrometer scale might correlate with the body size of an animal at the macroscopic level. By inputting 722 skin cell images and corresponding size data for each zebrafish larva into machine learning models, we determined that the Vision Transformer (ViT) with a random cropping and voting strategy was able to achieve high predictive performance (F-score of 0.91). Remarkably, analyzing as few as 27 skin cells within a single image of 0.01 mm ² was sufficient to predict the individual’s overall size, ranging from 0.9 to 3.1 mm ² . Using a gradient-weighted class activation map (Grad-CAM), we further identified the cellular features influencing the model’s decisions. These findings provide a proof-of-concept that macroscopic organismic information may be de-encrypted from a snapshot of only a few dozen cells using deep learning approaches.

BioStudies – one package for all the data supporting a study

Summary: Flooding is a widespread natural disaster that causes tremendous yield losses of global food production. Rice is the only cereal capable of growing in aquatic environments. Direct seeding by which seedlings grow underwater is an important cultivation method for reducing rice production cost. Hypoxic germination tolerance and root growth in waterlogged soil are key traits for rice adaptability to flooded environments. Alternative oxidase (AOX) is a non‐ATP‐producing terminal oxidase in the plant mitochondrial electron transport chain, but its role in hypoxia tolerance had been unclear. We have discovered that AOX1a is necessary and sufficient to promote germination/coleoptile elongation and root development in rice under flooding/hypoxia. Hypoxia enhances endogenous H₂O₂ accumulation, and H₂O₂ in turn activates an ensemble of regulatory genes including AOX1a to facilitate the conversion of deleterious reactive oxygen species to H₂O₂ in rice under hypoxia. We show that AOX1a and H₂O₂ act interdependently to coordinate three key downstream events, that is, glycolysis/fermentation for minimal ATP production, root aerenchyma development and lateral root emergence under hypoxia. Moreover, we reveal that ectopic AOX1a expression promotes vigorous root and plant growth, and increases grain yield under regular irrigation conditions. Our discoveries provide new insights into a unique sensor–second messenger pair in which AOX1a acts as the sensor perceiving low oxygen tension, while H₂O₂ accumulation serves as the second messenger triggering downstream root development in rice against hypoxia stress. This work also reveals AOX1a genetic manipulation and H₂O₂ pretreatment as potential targets for improving flooding tolerance in rice and other crops.

Cellular bioenergetics and mitochondrial dynamics are crucial for the secretion of insulin by pancreatic beta cells in response to elevated levels of blood glucose. To elucidate the interactions between energy production and mitochondrial fission/fusion dynamics, we combine live-cell mitochondria imaging with biophysical-based modeling and graph-based network analysis. The aim is to determine the mechanism that regulates mitochondrial morphology and balances metabolic demands in pancreatic beta cells. A minimalistic differential equation-based model for beta cells is constructed that includes glycolysis, oxidative phosphorylation, calcium dynamics, and fission/fusion dynamics, with ATP synthase flux and proton leak flux as main regulators of mitochondrial dynamics. The model shows that mitochondrial fission occurs in response to hyperglycemia, starvation, ATP synthase inhibition, uncoupling, and diabetic conditions, in which the rate of proton leakage exceeds the rate of mitochondrial ATP synthesis. Under these metabolic challenges, the propensities of tip-to-tip fusion events simulated from the microscopy images of the mitochondrial networks are lower than those in the control group and prevent the formation of mitochondrial networks. The study provides a quantitative framework that couples bioenergetic regulation with mitochondrial dynamics, offering insights into how mitochondria adapt to metabolic challenges.