BMC Bioinformatics最新文献

Background: Virtual Screening (VS) has become an essential tool in drug discovery, enabling the rapid and cost-effective identification of potential bioactive molecules. Among recent advancements, Graph Neural Networks (GNNs) have gained prominence for their ability to model complex molecular structures using graph-based representations. However, the integration of explainable methods to elucidate the specific contributions of molecular substructures to biological activity remains a significant challenge. This limitation hampers both the interpretability of predictive models and the rational design of novel therapeutics.

Results: We trained 20 GNN models on a dataset of small molecules with the goal of predicting their activity on 20 distinct protein targets from the Kinase family. These classifiers achieved state-of-the-art performance in virtual screening tasks, demonstrating high accuracy and robustness on different targets. Building upon these models, we implemented the Hierarchical Grad-CAM graph Explainer (HGE) framework, enabling an in-depth analysis of the molecular moieties driving protein-ligand binding stabilization. HGE exploits Grad-CAM explanations at the atom, ring, and whole-molecule levels, leveraging the message-passing mechanism to highlight the most relevant chemical moieties. Validation against experimental data from the literature confirmed the ability of the explainer to recognize a molecular pattern of drugs and correctly annotate them to the known target.

Conclusions: Our approach may represent a valid support to shorten both the screening and the hit discovery process. Detailed knowledge of the molecular substructures that play a role in the binding process can help the computational chemist to gain insights into the structure optimization, as well as in drug repurposing tasks.

Understanding the interplay between diseases and genes is crucial for gaining deeper insights into disease mechanisms and optimizing therapeutic strategies. In recent years, various computational methods have been developed to uncover potential disease-gene associations. However, existing computational approaches for disease-gene association prediction still face two major limitations. First, most current studies focus on constructing complex heterogeneous graphs using multi-dimensional biological entity relationships, while overlooking critical latent interaction patterns, namely, disease neighbor interactions and gene neighbor interactions-which are more valuable for association prediction. Second, in self-supervised learning (SSL), the presence of noise in auxiliary tasks commonly affects the accurate modeling of diseases and genes. In this study, we propose a novel denoising method for disease-gene association prediction, termed DGSL. To address the first issue, we utilize bipartite graphs corresponding to diseases and genes to derive disease-disease and gene-gene similarities, and further construct disease and gene interaction graphs to capture the latent interaction patterns. To tackle the second challenge, we implement cross-view denoising through adaptive semantic alignment in the embedding space, while preserving useful neighbor interactions. Extensive experiments on benchmark datasets demonstrate the effectiveness of our method.

Background: An important challenge in flow cytometry (FCM) data analysis is making comparisons of corresponding cell populations across multiple FCM samples. An interesting solution is creating a statistical mixture model for multiple samples simultaneously, as such a multi-sample model can characterize a heterogeneous set of samples, and facilitates direct comparison of cell populations across the data samples. The multi-sample approach to statistical mixture modeling has been explored in a number of reports, mostly within a Bayesian framework and with high computational complexity. Although these approaches are effective, they are also computationally demanding, and therefore do not relate well to the requirement of scalability, which is essential in the multi-sample setting. This limits their utility in the analysis of large sets of large FCM samples.

Results: We show that basic Gaussian mixture models can be extended to large data sets consisting of multiple samples, using a computationally efficient implementation of the expectation-maximization algorithm. We show that the multi-sample Gaussian mixture model (MSGMM) is competitive with other models, in both rare cell detection and sample classification accuracy. This allows us to further explore the utility of MSGMMs in the analysis of heterogeneous sets of samples. We demonstrate how simple heuristics on MSGMM model output can directly reveal structural patterns in a collection of FCM samples.

Conclusions: We recover the efficiency and utility of the basic MSGMM which underlies more complex and non-parametric Bayesian hierarchical mixture models. The possibility of fitting GMMs to large sets of FCM samples provides opportunities for the discovery of associations between sample composition and sample meta-data such as treatment responses and clinical outcomes.

Background: In this paper, we introduce an image analysis approach for spatiotemporal segmentation, quantification, and visualization of movement or contraction patterns in 2D+t and 3D+t microscopy recordings of biological tissues. The development of this pipeline was motivated by the observation of contraction waves in the extra-embryonic membranes of the red flour beetle Tribolium castaneum. These contraction waves are a novel finding, whose origin and function are not yet understood. The objective of the proposed approach is to analyze the dynamics of the extra-embryonic membranes in order to provide quantitative evidence for the existence of contraction waves during late stages of embryonic development.

Results: We apply the pipeline to live-imaging data of Tribolium embryonic development recorded with light-sheet fluorescence microscopy. The proposed pipeline integrates particle image velocimetry (PIV) for quantitative movement analysis, surface detection, tissue cartography, and algorithmic identification of characteristic movement dynamics. We demonstrate that our approach reliably and efficiently detects contraction waves in both 2D+t and 3D+t recordings and enables automated quantitative analyses, such as measuring the area involved in contractile behavior, wave duration and frequency, spatiotemporal location of the contractile regions, and their relation to the underlying velocity distribution.

Conclusions: The pipeline will be employed in future work to conduct a large-scale characterization and quantification of contraction wave behavior in Tribolium development and can be readily adapted for the identification and segmentation of characteristic tissue dynamics in other biological systems.

Multimodal visualizations are essential for identifying and interpreting complex relationships in diverse, high-dimensional biological datasets. However, existing visualization tools often lack native capabilities for embedding explicit statistical and computational annotations, hindering effective quantitative interpretation. We introduce MultiModalGraphics, an R package designed specifically for creating annotated scatterplots and heatmaps of multi-omics and high-dimensional biological data. The package allows seamless embedding of statistical summaries such as fold-changes, p-values, q-values, and standard deviations, facilitating direct quantitative comparisons. MultiModalGraphics interoperates with Bioconductor packages including MultiAssayExperiment, limma, voom, and iClusterPlus, streamlining workflows from data preprocessing and differential expression analysis to visualization. Case studies on three distinct real-world multimodal datasets illustrate its practical utility. Source code, documentation, and example datasets are available via GitHub ( https://github.com/famanalytics0/MultiModalGraphics ) and under review for inclusion into Bioconductor.

Background: The annotation of protein functions constitutes a key connection between genetic sequences, molecular conformations, and biochemical roles, driving progress in biomedical studies. Traditional experimental methods are time-consuming and resource-intensive, making it difficult to meet the demand for functional annotation of a vast number of proteins in the post-genomic era. The development of high-throughput sequencing technology has generated a large amount of protein-protein interaction (PPI) data. Prediction methods based on network topology have attracted attention due to their high efficiency and interpretability. The FSWeight algorithm calculates functional similarity by evaluating the commonality of second-order neighbors of proteins. However, it has limitations in terms of insufficient local information and a limited global perspective.

Results: In this study, we propose the topology-aware functional similarity (TAFS) framework, which integrates local neighborhood information with global topological information. A distance-dependent functional attenuation factor γ is introduced to dynamically adjust the weights of distant nodes, and a bidirectional joint co-function probability model is constructed. Experiments show that TAFS outperforms traditional baseline methods in both single-species and cross-species evaluations.

Conclusion: TAFS significantly improves prediction accuracy and interpretability through refined topological modeling, providing new insights for functional inference in complex biological networks.

The sequences and structures of the 5' mRNA regions in prokaryotes significantly impact transcript stability and translation efficiency at the post-transcriptional level. However, the structure-function relationship of the post-transcription regulation region remains elusive. Here, we present a dual-channel neural network integrating sequence-structure features to predict post-transcriptional gene expression in Escherichia coli. Our model combines Word2Vec and K-mer encoding for initial sequence representation, followed by parallel feature extraction via CNN and BiLSTM. An attention mechanism dynamically prioritizes critical elements within both channels. The feature vectors from these modules are concatenated and fed into a fully connected network for final prediction. Evaluated on randomized train-test splits, the model demonstrates robust performance in classifying expression levels based on 5' mRNA regions and can identify post-transcriptional regulation regions with high translational efficiency, with accuracy reaching 93%. This framework provides a computational tool for optimizing synthetic biology designs by linking sequence architecture to expression outcomes, enhancing the efficiency of biological sequence design.

Background: Vasculature is an essential part of all tissues and organs and is involved in a wide range of different diseases. However, available software for blood vessel image analysis is often limited: Some only process two-dimensional data, others lack batch processing, putting a time burden on the user, while still others require tightly defined culturing methods and experimental conditions. This highlights the need for software that has the ability to batch process three-dimensional image data and requires few and simple experimental preparation steps.

Results: We present VESNA, a Fiji (ImageJ) macro for automated segmentation and skeletonization of three-dimensional fluorescence images, enabling quantitative vascular network analysis. It requires only basic experimental preparation, making it highly adaptable to a wide range of possible applications across experimental goals and different tissue culturing methods. The macro's potential is demonstrated on a range of different image data sets, from organoids with varying sizes, network complexities, and growth conditions, to expanding to other 3D tissue culturing methods, with an example of hydrogel-based cultures.

Conclusions: With its ability to process large amounts of 3D image data and its flexibility across experimental conditions, VESNA fulfills previously unmet needs in image processing of vascular structures and can be a valuable tool for a variety of experimental setups around three-dimensional vasculature, such as drug screening, research in tissue development and disease mechanisms.

Allele-specific expression (ASE) analyses from RNA-Seq data provide quantitative insights into genomic imprinting and the genetic variants that affect transcription. Robust ASE analysis requires the integration of multiple computational steps, including read alignment, read counting, data visualization, and statistical testing-this complexity creates challenges for reproducibility, scalability, and ease of use. Here, we present ASE Toolkit (ASET), an end-to-end pipeline that streamlines SNP-level ASE data generation, visualization, and testing for parent-of-origin (PofO) effect. ASET includes a modular pipeline built with Nextflow for ASE quantification from short-read transcriptome sequencing reads, an R library for data visualization, and a Julia script for PofO testing. ASET performs comprehensive read quality control, SNP-tolerant alignment to reference genomes, read counting with allele and strand resolution, annotation with genes and exons, and estimation of contamination. In sum, ASET provides a complete and easy-to-use solution for molecular and biomedical scientists to identify and interpret patterns of ASE from RNA-Seq data.