Explore the vast range of titles available.

Background Advancements in single-cell RNA sequencing have enabled the analysis of millions of cells, but integrating such data across samples and methods while mitigating batch effects remains challenging. Deep learning approaches address this by learning biologically conserved gene expression representations, yet systematic benchmarking of loss functions and integration performance is lacking. Results We evaluate 16 integration methods using a unified variational autoencoder framework, incorporating batch and cell-type information. Results reveal limitations in the single-cell integration benchmarking index (scIB) for preserving intra-cell-type information. To address this, we introduce a correlation-based loss function and enhance benchmarking metrics to better capture biological conservation. Using cell annotations from lung and breast atlases, our approach improves biological signal preservation. We propose a refined integration framework, scIB-E, and metrics that provide deeper insights into the integration process and offer guidance for advanced developments in integrating increasingly complex single-cell data. Conclusions This benchmark highlights the potential of deep learning-based approaches for single-cell data integration, emphasizing the importance of biologically informed metrics and improved benchmarking strategies.

This paper firstly designs a five-dimensional model of learners’ characteristics (learners’ English reading ability, cognitive style, learning goal, learning situation, and learning effect) and a three-dimensional model of English reading resources’ characteristics (question types, topics, and difficulty of resources) in a fragmented learning environment through literature research. At the same time, to make the learning resources meet the characteristics of fragmented learning time and space, the English Level 4 reading resources are reasonably designed and segmented to adapt to the needs of learners’ mobile fragmented learning. Then, combined with machine learning algorithms, an adaptive recommendation model of learning resources in English fragmented reading is constructed. The algorithm-based adaptive recommendation algorithm for English fragmented reading resources is designed. Based on the generated decision trees, the expression rules are parsed to achieve adaptive pushing of resources. The results of this study show that adaptive recommendation of learning resources in English fragmented reading can help teachers to develop future resource recommendation strategies through effective data collection to adaptively push resources that are close to learners’ individual needs. The use of mobile by English learners to learn to read in a fragmented learning context enables targeted training in weak areas of English reading, thus enhancing different aspects of learners’ reading skills.

Targeted protein degradation is a powerful tool for biological research, cell therapy, and synthetic biology. However, conventional methods often depend on pre-fused degrons or chemical degraders, limiting their wider applications. Here we develop a guided protein labeling and degradation system (GPlad) in Escherichia coli , using de novo designed guide proteins and arginine kinase (McsB) for precise degradation of various proteins, including fluorescent proteins, metabolic enzymes, and human proteins. We expand GPlad into versatile tools such as antiGPlad, OptoGPlad, and GPTAC, enabling reversible inhibition, optogenetic regulation, and biological chimerization. The combination of GPlad and antiGPlad allows for programmable circuit construction, including ON/OFF switches, signal amplifiers, and oscillators. OptoGPlad-mediated degradation of MutH accelerates E. coli evolution under protocatechuic acid stress, reducing the required generations from 220 to 100. GPTAC-mediated degradation of AroE enhanced the titer of 3-dehydroshikimic acid to 92.6 g/L, a 23.8% improvement over the conventional CRISPR interference method. We provide a tunable, plug-and-play strategy for straightforward protein degradation without the need for pre-fusion, with substantial implications for synthetic biology and metabolic engineering. Targeted protein degradation in bacteria typically requires fusion with tags or chemical degraders. Here, authors developed GPlad, a tunable system using designed guide proteins and arginine kinase to degrade diverse proteins in E. coli without the need for exogenous degraders or protein fusions.

A fast hyperspectral image classification algorithm with strong noise robustness is proposed in this paper, aiming at the hyperspectral image classification problems under noise interference. Based on the Fast 3D Convolutional Neural Network (Fast-3DCNN), this algorithm enables the classification model to have good tolerance for various types of noise by using a Minimum Noise Fraction (MNF) as dimensionality reduction module for hyperspectral image input data. In addition, by introducing lightweight hybrid attention modules with the spatial and the channel information, the deep features extracted by the Convolutional Neural Network are further refined, ensuring that the model has high classification accuracy. Public dataset experiments have shown that compared to traditional methods, the MNF in this algorithm reduces the dimensionality of input spectral data, preserves information with higher signal-to-noise ratio(SNR) in the spectral bands, and aggregates spectral features into class feature vectors, greatly improving the noise robustness of the model. At the same time, based on a lightweight spectral–spatial hybrid attention mechanism, combined with fewer spectral dimensions, the model effectively avoids overfitting. With less loss in model training speed, it achieved better classification accuracy in small-scale training sample experiments, fully demonstrating the good generalization ability of this algorithm.

Deep eutectic solvents (DESs) have been widely used to capture CO2 in recent years. Understanding CO2 mechanisms by DESs is crucial to the design of efficient DESs for carbon capture. In this work, we studied the CO2 absorption mechanism by DESs based on ethylene glycol (EG) and protic ionic liquid ([MEAH][Im]), formed by monoethanolamine (MEA) with imidazole (Im). The interactions between CO2 and DESs [MEAH][Im]-EG (1:3) are investigated thoroughly by applying 1H and 13 C nuclear magnetic resonance (NMR), 2-D NMR, and Fourier-transform infrared (FTIR) techniques. Surprisingly, the results indicate that CO2 not only binds to the amine group of MEA but also reacts with the deprotonated EG, yielding carbamate and carbonate species, respectively. The reaction mechanism between CO2 and DESs is proposed, which includes two pathways. One pathway is the deprotonation of the [MEAH]+ cation by the [Im]− anion, resulting in the formation of neutral molecule MEA, which then reacts with CO2 to form a carbamate species. In the other pathway, EG is deprotonated by the [Im]−, and then the deprotonated EG, HO-CH2-CH2-O−, binds with CO2 to form a carbonate species. The absorption mechanism found by this work is different from those of other DESs formed by protic ionic liquids and EG, and we believe the new insights into the interactions between CO2 and DESs will be beneficial to the design and applications of DESs for carbon capture in the future.

Deep eutectic solvents (DESs) have been widely used to capture CO in recent years. Understanding CO mechanisms by DESs is crucial to the design of efficient DESs for carbon capture. In this work, we studied the CO absorption mechanism by DESs based on ethylene glycol (EG) and protic ionic liquid ([MEAH][Im]), formed by monoethanolamine (MEA) with imidazole (Im). The interactions between CO and DESs [MEAH][Im]-EG (1:3) are investigated thoroughly by applying H and C nuclear magnetic resonance (NMR), 2-D NMR, and Fourier-transform infrared (FTIR) techniques. Surprisingly, the results indicate that CO not only binds to the amine group of MEA but also reacts with the deprotonated EG, yielding carbamate and carbonate species, respectively. The reaction mechanism between CO and DESs is proposed, which includes two pathways. One pathway is the deprotonation of the [MEAH] cation by the [Im] anion, resulting in the formation of neutral molecule MEA, which then reacts with CO to form a carbamate species. In the other pathway, EG is deprotonated by the [Im] , and then the deprotonated EG, HO-CH -CH -O , binds with CO to form a carbonate species. The absorption mechanism found by this work is different from those of other DESs formed by protic ionic liquids and EG, and we believe the new insights into the interactions between CO and DESs will be beneficial to the design and applications of DESs for carbon capture in the future.

Multicellular organisms require intercellular and intracellular signaling to coordinately regulate different cell functions. Although many methods of cell-cell communication (CCC) inference have been developed, they seldom account for both the intracellular signaling responses and global spatial information. The recent advancement of spatial transcriptomics (ST) provides unprecedented opportunities to better decipher CCC signaling and functioning. In this paper, we propose an ST-based multilayer network method, stMLnet, for inferring spatial intercellular communication and multilayer signaling regulations by quantifying distance-weighted ligand–receptor signaling activity based on diffusion and mass action models and mapping it to intracellular targets. We benchmark stMLnet with existing methods using simulation data and 8 real datasets of cell type-specific perturbations. Furthermore, we demonstrate the applicability of stMLnet on six ST datasets acquired with four different technologies (e.g., seqFISH+, Slide-seq v2, MERFIS and Visium), showing its effectiveness and reliability on ST data with varying spatial resolutions and gene coverages. Finally, stMLnet identifies positive feedback circuits between alveolar epithelial cells, macrophages, and monocytes via multilayer signaling pathways within a COVID-19 microenvironment. Our proposed method provides an effective tool for predicting multilayer signaling regulations between interacting cells, which can advance the mechanistic and functional understanding of spatial CCCs.

Advancements in single-cell RNA sequencing (scRNA-seq) have enabled the analysis of millions of cells, but integrating such data across samples and methods while mitigating batch effects remains challenging. Deep learning approaches address this by learning biologically conserved gene expression representations, yet systematic benchmarking of loss functions and integration performance is lacking. This study evaluated 16 integration methods using a unified variational autoencoder framework, incorporating batch and cell-type information. Results revealed limitations in the single-cell integration benchmarking index (scIB) for preserving intra-cell-type information. To address this, we introduced a correlation-based loss function and enhanced benchmarking metrics to better capture biological conservation. Using annotations from the Human Lung Cell Atlas and Human Fetal Lung Cell Atlas, our approach improved biological signal preservation. This work highlights the need for biologically informed metrics in scRNA-seq integration and offers guidance for future deep learning developments.Competing Interest StatementThe authors have declared no competing interest.

A fundamental limitation of probabilistic deep learning is its predominant reliance on Gaussian priors. This simplistic assumption prevents models from accurately capturing the complex, non-Gaussian landscapes of natural data, particularly in demanding domains like complex biological data, severely hindering the fidelity of the model for scientific discovery. The physically-grounded Boltzmann distribution offers a more expressive alternative, but it is computationally intractable on classical computers. To date, quantum approaches have been hampered by the insufficient qubit scale and operational stability required for the iterative demands of deep learning. Here, we bridge this gap by introducing the Quantum Boltzmann Machine-Variational Autoencoder (QBM-VAE), a large-scale and long-time stable hybrid quantum-classical architecture. Our framework leverages a quantum processor for efficient sampling from the Boltzmann distribution, enabling its use as a powerful prior within a deep generative model. Applied to million-scale single-cell datasets from multiple sources, the QBM-VAE generates a latent space that better preserves complex biological structures, consistently outperforming conventional Gaussian-based deep learning models like VAE and SCVI in essential tasks such as omics data integration, cell-type classification, and trajectory inference. It also provides a typical example of introducing a physics priori into deep learning to drive the model to acquire scientific discovery capabilities that breaks through data limitations. This work provides the demonstration of a practical quantum advantage in deep learning on a large-scale scientific problem and offers a transferable blueprint for developing hybrid quantum AI models.

Electrocatalytic reduction of CO 2 into multicarbon (C 2+ ) products is a highly attractive route for CO 2 utilization; however, the yield of C 2+ products remains low because of the limited C 2+ selectivity at high CO 2 conversion rates. Here we report a fluorine-modified copper catalyst that exhibits an ultrahigh current density of 1.6 A cm −2 with a C 2+ (mainly ethylene and ethanol) Faradaic efficiency of 80% for electrocatalytic CO 2 reduction in a flow cell. The C 2– 4 selectivity reaches 85.8% at a single-pass yield of 16.5%. We show a hydrogen-assisted C–C coupling mechanism between adsorbed CHO intermediates for C 2+ formation. Fluorine enhances water activation, CO adsorption and hydrogenation of adsorbed CO to CHO intermediate that can readily undergo coupling. Our findings offer an opportunity to design highly active and selective CO 2 electroreduction catalysts with potential for practical application. Electrocatalytic reduction of CO 2 into multicarbon (C 2+ ) products is a highly attractive route for CO 2 utilization. Now, a fluorine-modified copper catalyst is shown to achieve current densities of 1.6 A cm −2 with a C 2+ Faradaic efficiency of 80% for electrocatalytic CO 2 reduction in a flow cell.