Wuhan University of Science & Technology

DNA-binding proteins (DBPs) are fundamental to many key cellular processes, possessing distinct binding domains, differential affinities for single- and double-stranded DNA structures, and playing roles in fundamental biological functions such as DNA replication and gene regulation. They are intimately linked to the pathological mechanisms of diseases like neurodegenerative disorders and cancers, making their prediction pivotal for unraveling protein function and disease mechanisms. However, conventional experimental techniques for DBP identification are temporally inefficient, labor-intensive, material-intensive and pricey. Existing DNA-binding protein prediction models either lack integration with pre-trained protein language models, primarily relying on manually constructed features, or despite utilizing pre-trained language models, fail to extract sufficiently effective information from the features generated by these models. Hence, a pressing imperative persists to devise robust computational frameworks capable of precise and efficient DBP delineation. In this study, we propose a novel deep learning-based method named CNNCaps-DBP for the accurate prediction of DBPs from primary sequence information. Our methodology incorporates the pre-trained protein language model ESM C and enhances the embeddings via an attention augmented convolution module. The extracted features are then passed through a hybrid deep learning framework consisting of Capsule network and MLP to construct the final predictive model. To optimize the model's training process, we applied a dynamic learning rate scheduler utilized for lessening the risk of premature convergence and enhance the robustness of the learning process. Experimental results show that CNNCaps-DBP significantly outperforms previous models in terms of predictive performance. To further validate the robustness of the proposed model, we evaluated it on additional independent datasets, where CNNCaps-DBP consistently outperformed state-of-the-art methods. In addition, we conducted two case studies to interpret the predictions of our model, which demonstrates the strong predictive capability for DBP identification. The source code used in this study is available at: https://github.com/YZYAlex/CNNCaps-DBP.

This paper investigates the fixed-time synchronization (FXTS) and prescribed-time synchronization (PSTS) problems of state-dependent switching neural networks (SDSNNs) with stochastic disturbances and impulsive effects. By leveraging the average impulsive interval, comparison principle, and interval matrix methodology, this study advances a novel analytical framework. Departing from conventional approaches, we reformulate stochastic disturbed and impulsive SDSNNs as interval-parameter systems through rigorous interval matrix transformation. Consequently, we derive some sufficient conditions in the form of linear matrix inequalities (LMIs) to ensure the realization of FXTS and PSTS. Since impulsive effects can potentially compromise synchronization stability, careful controller design becomes critical. To address this challenge, we develop a unified proportional integral (PI) control framework. Through proper adjustment of its control parameters, this framework enables the system to achieve both FXTS and PSTS. Moreover, by reasonably configuring the relationship between the impulsive intensity and the prescribed time, the synchronization performance can be balanced. Finally, we demonstrate the effectiveness of the theoretical results through two examples.

Curved carbon structure can modulate the interaction between the catalyst and the reaction intermediates and enhance the intrinsic activity of the catalyst. However, it remains a significant challenge to prepare curved carbon without surfactants or templates. Herein, we report a facile strategy to prepare curved carbon via mechanical shaking process with polydopamine (PDA) as a precursor. Then, uniformly dispersed Pt nanoparticles (Pt NPs) are planted in situ on the curved carbon structure from free-standing TiO₂ nanosheets (C-rPDA-Pt/TiO₂ NSs) to prepare ultralow Pt loading (13.4 μg cm^-2) catalysts for acidic hydrogen evolution reaction (HER). The mass activity of C-rPDA-Pt/TiO₂ NSs is tenfold higher than that of the planar carbon supported Pt catalyst (C-sPDA-Pt/TiO₂ NSs) and four times higher than that of 40 % commercial Pt/C catalyst at an applied voltage of -0.1 V. DFT calculations suggest that curved carbon structure effectively adjusts the electronic structure of Pt active sites, thereby achieving optimal Gibbs free energy for hydrogen evolution and enhancing HER performance.