A community effort to optimize sequence-based deep learning models of gene regulation

Modularity in hardware improves manufacturability of power electronics systems and reduces maintenance cost. By paralleling modular ac-dc converters, power capacity of SiC-based power conversion system can be easily expanded. However, it is questionable if any filter topology can be realized as a modular block considering not only electric and magnetic circuits, but also mechanical structure and N −1 redundancy. Considering interleaving, filters without any coupling between channels will lead to an over-design. Meanwhile, filters with magnetical coupling may suffer from complexity in mechanical structure, and redundancy. This work proposes a concept of modular filter building block (FiBB), and a configuration of coupled inductors and a boost inductor to realize FiBB. Secondary-windings of N coupled inductors are connected in series and form a separate loop similar to a current mirror. A single inductor rated for a single channel can be placed on this loop, and realize boost inductors for all channels. Circulating current can be attenuated for any N to similar level. Under a single channel fault, the faulty converter can be easily isolated from the loop, achieving N −1 redundancy. The proposed FiBB can be easily extended for any number of channels without a significant change in mechanical structure.

Crowdsourced challenges are powerful catalysts for advancing biomedical research and fostering community-driven innovation. The DREAM Challenges initiative, in collaboration with the Personalized Environment and Genes Study (PEGS), held a competition to spur the development of predictive models for hypercholesterolemia risk. Participants were tasked with integrating diverse data, including environmental exposures, whole-genome sequencing, and geospatial information, from a large and diverse cohort to classify hypercholesterolemia and generate novel insights. The top-performing models, which primarily leveraged gradient boosting and random forest classifiers, demonstrated strong predictive performance, outperforming traditional polygenic scores (PGS). In Challenge Task 1, models from the two top-performing teams substantially outperformed the Challenge benchmark dataset using only PGS (AUROC = 0.7358), with AUROCs of 0.7933 and 0.7919, respectively. Beyond prediction, these models highlight the significant value of self-reported health and environmental exposure data, revealing them as informative predictors of hypercholesterolemia risk that complement established clinical and genetic factors. This paper details the design of the challenge, presents the top-performing models, and highlights the potential of integrative, multimodal approaches for understanding and predicting complex human diseases.

Does understanding brain function require characterizing every last synapse? The answer depends on the meaning of ‘understanding’. Mapping the full connectome of a whole mammalian brain with electron or super-resolution light microscopy is surely a daunting challenge (Lu, 2011). The paper in this issue of The Journal of Physiology by Ramaswamy et al. (2012) shows that simple stochastic selection of contacts among axo-dendritic spatial proximities captures the statistics of somatically recorded physiological signals. Digital reconstructions of neuronal morphology were randomly distributed in a simulated 3D cortical volume. Virtual synapses were then assigned to incidental axo-dendritic overlaps, modelling kinetics and amplitudes based on experimental measures. The resulting somatic EPSPs recorded in silico reproduced the distributions observed in vitro. It is too early to say how this finding will generalize across cortical regions, layers, and pre- and postsynaptic cell types. Most importantly, it remains unknown if neuronal morphology is sufficient to specify all other essential aspects of synaptic computation, such as efficacy (Komendantov & Ascoli, 2009), but also integration, homeostasis, plasticity and metabotropic cascades. This new report nevertheless lends support to the notion that detailed circuitry may reflect individual experience rather than general coding principles (Kalisman et al. 2005). Though experience may significantly shape circuitry, the adult cortex is not a genuine ‘blank slate’. It cannot adapt its wiring to any arbitrary pattern, because its axonal and dendritic arbours do constrain the formation of synapses, powerfully gating what can be learned. The contribution of this paper lies in showing that the postsynaptic potentials reaching the soma are statistically determined by the morphology of identified neuronal classes rather than by the specific synapse locations. Thus, comprehensive knowledge of every single synapse (what Americans might call ‘no synapse left behind’) may not be strictly necessary to reverse-engineer the brain. Instead, an anatomically realistic virtual network might be simulated by generating spatial distributions of appropriately shaped, packed and oriented neuronal forests re-sampled from large amounts of three-dimensional digital reconstructions at the light microscopy level (Kozloski, 2011). A new perspective emerges from the evidence accumulating in this and several other studies. Two distinct levels of connectomics can be recognized. One is that of actual synapses, with vesicle release, neurotransmitter receptors, etc. The other is that of potential contacts, consisting of spatial proximities of axonal and dendritic branches (Stepanyants & Chklovskii, 2005). It would be too reductive, and often incorrect, to consider synapses simply as a random subset of axo-dendritic proximities or to view axo-dendritic proximities as synaptic predictors (Peter's rule: Braitenberg and Schüz, 1998). Actual and potential connectivity represent two different physical entities, each with a separate and unique function. Their exact cognitive correlates are yet to be discovered, but it is tempting to speculate. Actual circuits might reflect knowledge, while potential circuits might relate to the capability to acquire new knowledge (given the right experience). The full specification of both actual and potential connectivity may characterize an individual and that individual's specific history, while the overall population statistics of those measures may quantify an attribute of the species. Scaling out further, the modular building blocks or repeating motifs of neural networks at both the actual and potential levels of description could reflect underlying computing mechanisms. The general patterns of dendritic and axonal morphology in each neuron class take shape during development. Different brain regions mature asynchronously and display critical windows for rewiring at dramatically disparate ages. Certain parts of the brain, such as the prefrontal cortex and especially the hippocampus, maintain high structural plasticity throughout adulthood. In these regions, spines twitch in and out of dendrites in a matter of minutes, boutons crawl up and down their axons in an activity- and experience-dependent manner, and newborn neurons are continuously embedded in the functional blueprint. These dynamics continuously rewire the actual connectome within the more stable potential connectivity map. Thus, in hippocampal and similarly plastic cortices, potential connections might constitute more relevant neural correlates of information processing than actual synapses. It is also intriguing that synaptic invariance results particularly from the morphological diversity of neuronal arbours. Although the general branching patterns (Sholl distributions, etc.) are considered revealing signatures of a given neuronal class, no two neurons are ever exactly alike. The actual meandering and locations of axons and dendrites constitute the real constraints of (and potential for) synaptic connectivity. These specific tree configurations also change, although at a slower time scale than synapses, possibly corresponding to long-term learning processes, such as the progressive adaptation to new contexts or environments. By providing fixed links, actual synapses delimit the extent of this branch-level plasticity, perhaps reflecting the common experience that old knowledge sometimes hampers the acquisition of new knowledge. The article also illustrates important methodological considerations that are taken for granted in physics, but are still too often underappreciated in neuroscience. Modelling is an essential element in the process of discovery, along with experiments and theory. Whereas experiments can disprove theoretical assumptions by falsifying hypotheses, computer simulations enable the demonstration that a theoretical explanation is complete, by showing that the postulated mechanism is sufficient to reproduce the observed phenomenon. Agile theoretical developments are frequently better aided by flexible and extensive exploration of models than by slow, costly, invasive and technologically limited experiments. Theory indicates how experimental measures should constrain and validate computer simulations, and how those simulations can help interpret empirical results, integrate knowledge and design new experiments. Neuroanatomy is particularly needy in these regards, because the complexity of human brain connectivity is so massive that data collection alone will not suffice to explain its function. Although only a few years old, computational neuroanatomy has rapidly advanced into a powerful approach with mature resources (e.g. NeuroMorpho.Org) and a thriving research community. I am grateful to Dr Rebecca F. Goldin for constructive feedback on an earlier version of this Perspective. Partial support was provided by the National Institutes of Health (R01 NS39600) and the Office of Naval Research (MURI N00014-10-1-0198).

Recent progress on machine learning techniques for flow field optimization toward performance and commercial scale proton exchange membrane fuel cell: A systematic review

We assembled eight cofacial porphyrin prisms using MTPyP (M = Co(II) or Zn(II), TPyP = 4-tetrapyridylporphyrin) and functionalized ruthenium-based "molecular clips" using coordination-driven self-assembly. Our approach allows for the rapid synthesis of these architectures in isolated yields as high as 98% for the assembly step. Structural and reactivity studies provided a deeper understanding of the role of the building blocks on the oxygen reduction reaction (ORR). Catalytic efficacy was probed by using cyclic and hydrodynamic voltammetry on heterogeneous catalyst inks in aqueous media. The reported prisms showed outstanding selectivity (>98%) for the kinetically hindered 4e-/4H+ reduction of O2 to H2O over the kinetically more accessible 2e-/2H+ reduction to H2O2. Furthermore, we have demonstrated significant cofacial enhancement in the observed catalytic rate constant ks (∼5 orders of magnitude) over the mononuclear analogue. We conclude that the steric bulk of the clip plays an important role in the structural dynamics of these prisms, which in turn modulates the ORR reactivity with respect to selectivity and kinetics.

Today many peer-to-peer (P2P) simulation frameworks feature a variety of recent years' research outcomes as modular building blocks, allowing others to easily reuse these blocks in further simulations and approach more advanced issues more rapidly. However, the efforts in the field of P2P load balan

Printable organic bioelectronics provide a fast and cost-effective approach for the fabrication of novel biodevices, while the general challenge is to achieve optimized reaction kinetics at multiphase boundaries between biomolecules and electrodes. Here, we present an entirely new concept based on a modular approach for the construction of heterostructured bioelectronic interfaces by using tailored functional "biological microparticles" combined with "transducer microparticles" as modular building blocks. This approach offers high versatility for the design and fabrication of bioelectrodes with a variety of forms of interparticle spatial organization, from layered-structures to more advance bulk heterostructured architectures. The heterostructured biocatalytic electrodes delivered twice the reaction rate and a six-fold increase in the effective diffusion kinetics in response to a catalytic model using glucose as the substrate, together with the advantage of shortened diffusion paths for reactants between multiple interparticle junctions and large active particle surface. The consequent benefits of this improved performance combined with the simple means of mass production are of major significance for the emerging printed electronics industry.

Efficient prototyping of a large complex system can be significantly facilitated by the use of a flexible and versatile physical platform where both new hardware and software components can readily be implemented and tightly integrated in a timely manner. Towards this end, we have developed the 120 130 mm QFDB board and associated firmware, including the system software environment. We developed a large system based on this advanced dense and modular building block. The QFDB features 4 interconnected Xilinx Zynq Ultrascale+ devices, each one consisting of an ARM-based subsystem tightly coupled with reconfigurable logic. Each Zynq Ultrascale+ is connected to 16 GB of DDR4 memory. In addition, one Zynq provides storage through an M.2 Solid State Disk (SSD). In this paper, we present the design and the implementation of this board, as well as the software environment for board operation. Moreover, we describe a 10 Gb Ethernet communication infrastructure for interconnecting multiple boards together. Finally, we highlight the impact of this board on a number of ongoing research activities that leverage the QFDB versatility, both as a largescale prototyping system for HPC solutions, and as a host for the development of FPGA integration techniques.

13 - Stem cell technology for in vitro bone tissue engineering

The accurate estimation of forest aboveground biomass (AGB) is essential for effective forest resource management and carbon stock assessment. However, the estimation accuracy of forest AGB is often constrained by scarce in situ measurements and the limitations of using a single data source or retrieval model. This study proposes a multi-source data integration framework using Sentinel-1 (S-1) and Sentinel-2 (S-2) data along with eight predictive models (i.e., multiple linear regression—MLR; Elastic-Net; support vector regression (with a linear kernel and polynomial kernel); k-nearest neighbor; back-propagation neural network—BPNN; random forest—RF; and gradient-boosting tree—GBT). With airborne light detection and ranging (LiDAR)-derived AGB as a reference, a three-stage optimization strategy was developed, including stepwise feature selection (SFS), hyperparameter optimization, and multi-decision vector fusion (MDVF) model construction. Initially, the optimal feature subsets for each model were identified using SFS, followed by hyperparameter optimization through a grid search strategy. Finally, eight models were evaluated, and MDVF was implemented to integrate outputs from the top-performing models. The results revealed that LiDAR-derived AGB demonstrated a strong performance (R2 = 0.89, RMSE = 20.27 Mg/ha, RMSEr = 15.90%), validating its effectiveness as a supplement to field measurements, particularly in subtropical forests where traditional inventories are challenging. SFS could adaptively select optimal variable subsets for different models, effectively alleviating multicollinearity. Satellite-based AGB estimation using the MDVF model yielded robust results (R2 = 0.652, RMSE = 31.063 Mg/ha, RMSEr = 20.4%) through the synergy of S-1 and S-2, with R2 increasing by 4.18–7.41% and the RMSE decreasing by 3.55–5.89% compared to the four top-performing models (BPNN, GBT, RF, MLR) in the second optimization stage. This study aims to provide a cost-effective and precise strategy for large-scale and spatially continuous forest AGB mapping, demonstrating the potential of integrating active and passive satellite imagery with airborne LiDAR to enhance AGB mapping accuracy and support further ecological monitoring and forest carbon accounting.

Accurate characterization of shaley-sand reservoirs remains a significant challenge in petroleum geophysics, where complex clay mineralogy often renders traditional evaluation methods unreliable. This study introduces an integrated, data-driven framework that synergizes numerical optimization and machine learning (ML) to accurately estimate formation water resistivity (Rw) and predict water saturation (Sw), overcoming the limitations of data scarcity. The workflow begins with rigorous preprocessing of well log data from 11 wells across the Norwegian North Sea and Egyptian Western Desert. First, we establish a robust, physically-constrained Rw by evaluating four optimization algorithms. The Powell and Nelder-Mead algorithms emerged as superior, demonstrating the ability to recover the true Rw from log data with low error (1×10-4 RMSE) against measured samples rapidly. This optimized Rw then serves as a high-quality "pseudo-core" label to generate a continuous Sw log for training a comprehensive suite of ML models, including ensemble methods (Random Forest, CatBoost, XGBoost) and neural networks (ANNs, LSTM), to predict Sw. The models demonstrated predictive accuracy, validated by a robust 5-fold cross-validation protocol. On the blind test wells, the top-performing models (LSTM, CatBoost , and XGBoost) achieved a coefficient of determination (R2) up to 0.944 with Mean Absolute Error (MAE) and Root Mean Squared Error (RMSE) as low as 0.03 and 0.050 respectively. The automated fusion of optimization-derived physics with ML-driven prediction marks a transformative step toward more reliable, data-centric petrophysical workflows. This integrated framework offers a significant enhancement in reservoir characterization, providing a cost-effective and scalable methodology that reduces reliance on expensive core analyses and improves the accuracy of hydrocarbon-in-place estimations.Supplementary InformationThe online version contains supplementary material available at 10.1038/s41598-026-36133-w.

Diabetes mellitus is a chronic metabolic disorder characterized by persistent high blood sugar levels due to insulin deficiencies, leading to complications such as diabetic foot ulcers (DFUs). DFUs are associated with high morbidity and a significant risk of amputation, necessitating precise monitoring and management. Manual measurement of ulcer areas is labor-intensive and error-prone, prompting the need for automated, computer-aided methods. Deep learning (DL) techniques have shown promise in this domain, enhancing the accuracy and efficiency of ulcer detection and segmentation. This study investigates various methodologies for DFU segmentation, focusing on advanced DL models. We propose FootSegONN, an EfficientNet-based encoder and Self-organized Operational Neural Network (Self-ONN) and Feature Pyramid Network-based decoder for foot ulcer segmentation, evaluated on a publicly available chronic wound dataset consisting of 1010 diabetic foot images. Self-ONNs address the limitations of conventional Convolutional neural networks by achieving ultimate heterogeneity and boosting network diversity while maintaining computational efficiency. To ensure robust validation, fivefold cross-validation was applied, and more than 10 different segmentation models were utilized. The STAPLE algorithm is employed to combine mask predictions from top-performing models, and a post-processing approach is investigated to enhance performance further. The proposed method achieved a state-of-the-art Dice score of 91.55%. Furthermore, the combined FootSegONN model achieved a Dice score of 88.97% on the external validation dataset. Gradient-weighted class activation map (Grad-CAM) visualization was utilized to assess the model’s interpretability.

The article presents neural network for recognition of driving strategies based on interactions between drivers in road traffic. It analyzes the architecture of the model implemented as a self-organizing map (SOM), consisting of a group of neural networks based on radial basis functions (RBF). It is a training model grounded in the biological foundations of artificial neural networks, in which the training set should consist exclusively of input vectors; wherein the network training algorithm adjusts itself the network's weights to obtain consistent output vectors (i.e. to make presenting sufficiently close input vectors result into the same outputs).\nThe article presents the results of using a new generation of the neural network developed by us, which includes an adaptive learning algorithm to reduce the effect of re-training (overfitting) and false recognition, as well as to improve the determination of the boundaries between clusters.\nThe aim of the research is to outline architecture and structure of the neural network model that allows recognizing strategical characteristics of driving and can identify strategies of interactions between vehicles (their drivers) in road traffic as well as identify behavioral patterns This paper considers driving strategies that characterize the interaction of dyads of vehicles (drivers) moving in road traffic. The research results show that the SOM RBF neural networks can recognize and classify types of interactions in road traffic based on modelling of the analysis of vehicle movement trajectories. Experimental results demonstrate the neural networks architecture and networks learning involving 400 iterations of streaming the training data representing 500 possible simulated interaction situations. This paper presents a novel neural network model for recognition of drivers' behaviour patterns and for classification of driving strategies into five general classes: (1) competition strategy, (2) contest strategy, (3) evasion strategy, (4) compromise strategy, and (5) active confrontation strategy. This neural network has demonstrated a high rate of recognition and concise clusterization of similar driving strategies. The key contribution of this paper: it proposes a neural network model based on Kohonen's Self-Organizing Map (SOM) for detecting drivers' behaviours from vehicle movement patterns - driving strategies - instead of monitoring driver's specific activities.

BackgroundImmune checkpoint inhibitors (ICIs) have transformed cancer therapy, yet their response rates remain modest, ranging from 20–40% across different cancer types.1 There exists a critical need for predictive tools to...

Recurrent neural network based hybrid model for reconstructing gene regulatory network