Interdisciplinary Sciences: Computational Life Sciences最新文献

Alzheimer's disease (AD) is a progressive neurodegenerative disorder with increasing prevalence among the ageing population, necessitating early and accurate diagnosis for effective disease management. In this study, we present a novel hybrid deep learning framework, AlzhiNet, that integrates both 2D convolutional neural networks (2D-CNNs) and 3D convolutional neural networks (3D-CNNs), along with a custom loss function and volumetric data augmentation, to enhance feature extraction and improve classification performance in AD diagnosis. According to extensive experiments, AlzhiNet outperforms standalone 2D and 3D models, highlighting the importance of combining these complementary representations of data. The depth and quality of 3D volumes derived from the augmented 2D slices also significantly influence the model's performance. The results indicate that carefully selecting weighting factors in hybrid predictions is imperative for achieving optimal results. Our framework has been validated on the magnetic resonance imaging (MRI) from Kaggle and MIRIAD datasets, obtaining accuracies of 98.9% and 99.99%, respectively, with an AUC of 100%. Furthermore, AlzhiNet was studied under a variety of perturbation scenarios on the Alzheimer's Kaggle dataset, including Gaussian noise, brightness, contrast, salt and pepper noise, color jitter, and occlusion. The results obtained show that AlzhiNet is more robust to perturbations than ResNet-18, making it an excellent choice for real-world applications. This approach represents a promising advancement in the early diagnosis and treatment planning for AD.

Identifying drug-target interaction (DTI) is crucial for drug discovery and repositioning. Biological identification of DTI is costly and time-consuming, so heterogeneous network methods for DTI prediction, which can speed up drug development, have drawn much attention. However, the scarcity of known drug-target pairs leads to an overwhelmingly sparse network structure. This sparsity makes it difficult for graph convolutional networks to fully capture the potential relationships between nodes during information transmission and feature extraction, thus challenging the prediction performance of DTI. Therefore, we propose a method, called VHGAE, based on a heterogeneous graph variational autoencoder to predict drug-target interactions. The VHGAE model integrates a wealth of prior knowledge from various sources related to drugs and targets to construct a heterogeneous network. To address the sparsity issue within the constructed heterogeneous network, first, the weighted k-nearest neighbor algorithm is used to densify the DTI network to increase the connectivity of the sparse network. Second, the weighted graph convolutional network within a variational graph autoencoder framework is employed to bolster the edge weights with limited connections. Third, a variational expectation maximization algorithm is introduced between the encoder and the decoder to recover potential relationships in the sparse network. The experimental results on two datasets show that the performance of VHGAE outperforms nine state-of-the-art DTI prediction methods, which suggests that the design of VHGAE in multi-source data fusion and sparse network processing plays an important role in improving prediction performance.

Single-cell RNA sequencing (scRNA-seq) plays a vital role in studying cellular heterogeneity and gene expression patterns. However, the sequencing dropout phenomena still pose a significant challenge. Genes with low expression levels may be misidentified as exhibiting zero expression owing to limitations in sequencing depth and technical noise. This results in increased data sparsity and compromises the accuracy of subsequent analyses. Thus, a novel method, MoDET (Dual-level Momentum Distillation Method with Extreme Thresholding), has been proposed. MoDET employs a label-guided model and an extreme threshold mechanism to enhance cellular representation learning. Experiments demonstrate that MoDET significantly improves clustering performance of the gene expression matrix, with enhancements ranging from 3% to 20% across seven real-world datasets. Cross-batch training and evaluation experiments demonstrated that MoDET effectively mitigates batch effects, achieving an average performance improvement of 5%-7%. Concurrently, it exhibits superior accuracy in identifying rare cell types, outperforming other methods by 3%-20%. Ablation studies confirm that the dual-level momentum distillation boosts performance by 4%-20%, and the extreme threshold mechanism adds an additional 2%-15% improvement. Interpretability analysis shows that the extreme threshold makes the model's decision-making process more transparent. Moreover, MoDET surpasses methods incorporating advanced modules, thereby demonstrating its efficacy in addressing the sparsity challenges inherent in scRNA-seq datasets. The compiled source codes are accessible at https://github.com/gladex/MoDET.

Anti-inflammatory peptides (AIPs) have emerged as potential therapeutic candidates for managing various inflammatory disorders, but their computational identification remains challenging. We propose AIP-TranLAC, a novel deep learning framework that integrates Transformer-based embedding, bidirectional long short-term memory (Bi-LSTM), multi-head attention, and convolutional neural network (CNN) to classify AIPs accurately. Our model achieves superior performance on benchmark and independent test datasets, demonstrating significant improvements over existing methods. The hybrid architecture effectively captures local and global sequence patterns, while interpretability analyses reveal critical amino acid residues. With robust performance on imbalanced data and open-source availability, AIP-TranLAC provides a powerful tool for accelerating therapeutic peptide discovery and inflammation research. For reproducibility purposes, we have released the codebase, trained models, and all supporting data on GitHub ( https://github.com/Renjingyi123/AIP-TranLAC ).

Proteins are fundamental to biological processes, mediating critical functions through precise molecular interactions. The rotational dynamics between ligand atoms and protein binding sites can significantly influence interaction efficacy by modifying spatial relationships. In our research, we present the PLAe (three-dimensional (3D) rotationally equivariant neural networks for predicting protein-ligand binding affinities) methodology. This novel model synergizes radial basis functions with e3nn networks to encapsulate the radial and angular dimensions of molecular features. Radial basis functions effectively measure interatomic distances, while e3nn-an advanced neural network utilizing spherical harmonics-maintains invariance under rotational and translational transformations. The Clebsch-Gordan coefficients are employed to integrate angular and atomic properties seamlessly. By merging radial basis and spherical harmonic elements with Clebsch-Gordan representations, our approach adeptly captures molecular rotational symmetries and interatomic interactions. The inclusion of an attention mechanism further refines the affinity predictions, ensuring a high level of precision. This integrative and sophisticated model sets a new benchmark to accurately predict protein-ligand binding affinities, leveraging intricate molecular details to enhance predictive performance.

Magnesium is an essential element involved in diverse life activities. The strong polarization and significant charge transfer effects pose challenges to the traditional fixed charge force fields. Here we establish the ABEEM/MM magnesium force field for proteins and aqueous solutions. The interaction potentials of magnesium with water and proteins are treated as the ABEEM/MM bonded model (ABEEM-BM) in the Morse potential function form. Based on quantum mechanical (QM) results, the related parameters are optimized and determined. The charge distributions of model molecules from ABEEM-BM and the ABEEM/MM nonbonded model (ABEEM-NBM) agree well with the QM results. The potential energy surfaces (PESs) for bond stretching and angle bending between magnesium and ligands by ABEEM-BM have a good consistency with those from QM. Molecular dynamics (MD) simulations of 40 aqueous magnesium protein segments are carried out using ABEEM-BM, ABEEM-NBM, OPLS-AA, AMBER99, and CHARMM22 force fields. The root mean square deviations (RMSDs) for bond length and angle by ABEEM-BM are 0.088 Å and 5.99°, respectively, which are smaller than those from the others. MD simulations of aqueous magnesium solutions are carried out using ABEEM-BM and ABEEM-NBM. The radial and angular distribution functions from ABEEM-BM reproduce the best structural properties, and the rate constant is 4.7 × 10⁵ s^- 1. Moreover, the dynamic changing picture of charge transfer and the coordination number (CN) during water exchange processes is presented by ABEEM model. The overall performance of ABEEM models is evidently better than those from fixed charge force fields.

The segmentation of brain tumor magnetic resonance imaging (MRI) plays a crucial role in assisting diagnosis, treatment planning, and disease progression evaluation. Convolutional neural networks (CNNs) and transformer-based methods have achieved significant progress due to their local and global feature extraction capabilities. However, similar to other medical image segmentation tasks, challenges remain in addressing issues such as blurred boundaries, small lesion volumes, and interwoven regions. General CNN and transformer approaches struggle to effectively resolve these issues. Therefore, a new multi-stage and adjacent-level feature integration network (MAI-Net) is introduced to overcome these challenges, thereby improving the overall segmentation accuracy. MAI-Net consists of dual-branch, multi-level structures and three innovative modules. The stage-level multi-scale feature extraction (SMFE) module focuses on capturing feature details from fine to coarse scales, improving detection of blurred edges and small lesions. The adjacent-level feature fusion (AFF) module facilitates information exchange across different levels, enhancing segmentation accuracy in complex regions as well as small volume lesions. Finally, the multi-stage feature fusion (MFF) module further integrates features from various levels to improve segmentation performance in complex regions. Extensive experiments on BraTS2020 and BraTS2021 datasets demonstrate that MAI-Net significantly outperforms existing methods in Dice and HD95 metrics. Furthermore, generalization experiments on a public ischemic stroke dataset confirm its robustness across different segmentation tasks. These results highlight the significant advantages of MAI-Net in addressing domain-specific challenges while maintaining strong generalization capabilities.

The identification of protein homologs in large databases is critical for biological advancements. Traditional methods, such as protein sequence alignment, often miss remote homologs. To address this limitation, we present the Basic Embedding Search Tool (BEST), a fast and sensitive approach that employs protein language models to create sequence embeddings enriched with evolutionary and structural information. Besides, we introduce a segmented distillation pruning technique to accelerate sequence encoding and develop a multi-layer acceleration structure to achieve a 4290.86-fold speedup in swift access and retrieval of dense vectors. Extensive experiments on real datasets demonstrate that BEST increases sensitivity by over 20% compared to prior methods while maintaining precision and recall. It operates 23.41 times faster than traditional tools like PSI-BLAST and 3.92 times faster than Foldseek, while also detecting homologous sequences that conventional methods miss. BEST and its open-access web server ( http://pm2s.cpolar.top/best1/ ) are poised to significantly aid enzyme mining and advance biological research. The code is publicly available at https://github.com/SkyTai-W/ProteinMiningEvaluator .

The latest progress in spatial transcriptomics has empowered scientists to investigate spatial heterogeneity with single-cell precision. A pivotal yet demanding aspect of spatial transcriptomics data analysis is cell type annotation. However, current methods exhibit limited performance as they are primarily designed for scRNA-seq data. Especially, these approaches often neglect spatial coordinate information and encounter challenges in identifying novel cell types. Here, we introduce SANNO, a novel approach that employs Optimal Transport (OT) to concurrently identify both known and novel cell types in spatially resolved single-cell data. Specifically, SANNO leverages a graph-Transformer module to model spatial coordinates and gene expression. This produces unified representations for both reference and query data. Building on this, SANNO employs a dual-strategy classifier. The first is an Unbalanced Optimal Transport (UOT) module that aligns query data with reference prototypes. The second is a self-supervised OT-based module that enhances global cluster separation and local cellular consistency, effectively eliminating batch effects. To further improve prediction accuracy, SANNO integrates an entropy-based re-weighted loss function. This significantly boosts the confidence of query cell predictions. Comprehensive experiments reveal that SANNO surpasses state-of-the-art techniques across both intra- and cross-spatial datasets, particularly in the identification of novel cell types. Additionally, SANNO demonstrates commendable performance in annotating cells within single-cell data, underscoring its potential as a versatile tool for cell annotation across single-cell and spatial transcriptomics datasets.