Frontiers in bioinformatics最新文献

Introduction: Citron Rho-interacting serine/threonine kinase (CIT) is a major cytosolic protein kinase essential for midbody organisation, abscission, and cytokinesis. Dysregulation and mutations in CIT are associated with multiple cancers and neurodevelopmental disorders, including microcephaly. Although global phosphoproteomic studies have identified more than 50 phosphosites in CIT, their functional relevance and the kinases regulating them remain largely unexplored.

Methods: To systematically investigate the phosphoregulation of CIT, we curated and integrated global phosphoproteomic datasets, along with their associated experimental conditions, to comprehensively catalogue phosphorylation events reported for CIT. To assess the functional significance of CIT, we examined proteins that were differentially co-regulated with its predominant phosphosite.

Results: Serine 440 (S440), located outside the kinase domain (representing over 55% of CIT-associated phospho-signalling events across 100 experimental conditions, including Enterovirus A71 infection, metformin, and interleukin-33), was identified as its predominant phosphosite. Motif analysis revealed the presence of a D(S/T)P/P(S/T)D motif recognised by the CIT kinase domain, suggesting S440 as a predicted autophosphorylation site. Co-phosphoregulation analysis identified 136 interacting proteins and 82 predicted substrates that were positively co-regulated with CIT_S440. The resulting phospho-regulatory network comprised essential cell cycle and DNA repair regulators, including MDC1 and TRIP12. Significantly, over 120 co-regulated phosphosites were functionally linked to DNA repair and cell cycle regulation. Aberrant phosphorylation of CIT_S440 observed across cancers of the breast, colon, and bladder suggests CIT_S440 as a potential onco-phosphosite critically involved in cellular checkpoint signalling.

Discussion: These findings suggest that CIT_S440 functions as a promising therapeutic target, and the phosphosite-centric regulatory network derived in this study could serve as a platform to evaluate its phosphosite-specific therapeutic interventions.

Background: As a driving force of the Fourth Industrial Revolution, deep learning methods have achieved significant success across various fields, including genetic and genomic studies. While individual-level genetic data is ideal for deep learning models, privacy concerns and data-sharing restrictions often limit its availability to researchers.

Methods: In this paper, we investigated the potential applications of deep learning models-including deep neural networks, convolutional neural networks, recurrent neural networks, and transformers-when only genetic summary data, such as linkage disequilibrium matrices, is available. The bootstrap method was used to approximate the test error. Simulation studies and real data analyses were conducted to compare the performance of deep learning methods in genetic risk prediction using individual-level genetic data versus genetic summary data.

Results: The test mean squared errors (MSEs) of most applied deep learning models are comparable when using individual-level data versus summary data.

Conclusion: Our results suggest that suitable deep learning methods could also serve as an alternative approach to predict disease related traits when only linkage disequilibrium matrices are available as input.

Platelet-rich plasma (PRP) has been widely applied in clinical practice for tissue repair and regeneration. Recent studies have reported that large amounts of extracellular vesicles (EVs) derived from PRP (PRP-EVs) are also involved in the functions of tissue repair and regeneration, except for the secreted growth factors. However, the relevant mechanisms of PRP-EVs remain unknown. In this study, we attempted to reveal the potential circular RNA (circRNA) mechanisms of PRP-EVs using high-throughput RNA sequencing (RNA-seq) technique and bioinformatics analysis. Six healthy donors were enrolled in this study, including three donors for the isolation of PRP-EVs and three donors for the isolation of EVs derived from blood plasma (plasma-EVs). As a result, we confirmed that PRP activation by thrombin could significantly promote the formation and secretion of EVs, particularly those with diameters ranging from 50 to 200 nm. Moreover, 144 circRNAs were altered in PRP-EVs with a fold change ≥ 2.0 and p-value ≤ 0.05. Among these, 89 circRNAs were upregulated, whereas 55 circRNAs were downregulated. Gene Ontology (GO), Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway, and circRNA-miRNA-mRNA interaction network analyses were performed to predict the potential roles of circRNAs in PRP-EVs. GO analysis indicated that these altered circRNAs might be related to the physiological processes of cell genesis and development. The pathways that were most strongly correlated with the biological functions of PRP-EVs were the transforming growth factor β (TGF-β) signaling pathway and HIF-1 signaling pathway. In addition, the expression levels of five selected circRNAs were verified through RT-qPCR. In conclusion, this is the first study to explain a novel potential mechanism of the biological functions of PRP-EVs in terms of the altered circRNAs. Taken together, our findings in this study may lay the groundwork for the clinical application of PRP-EVs and provide possible novel targets for further research.

Background: Biomedical knowledge graphs (KGs), such as the Data Distillery Knowledge Graph (DDKG), capture known relationships among entities (e.g., genes, diseases, proteins), providing valuable insights for research. However, these relationships are typically derived from prior studies, leaving potential unknown associations unexplored. Identifying such unknown associations, including previously unknown disease-associated genes, remains a critical challenge in bioinformatics and is crucial for advancing biomedical knowledge.

Methods: Traditional methods, such as linkage analysis and genome-wide association studies (GWAS), can be time-consuming and resource-intensive. This highlights the need for efficient computational approaches to identify or predict new genes using known disease-gene associations. Recently, network-based methods and KGs, enhanced by advances in machine learning (ML) frameworks, have emerged as promising tools for inferring these unexplored associations. Given the technical limitations of the Neo4j Graph Data Science (GDS) machine learning pipeline, we developed a novel machine learning pipeline called KG2ML (Knowledge Graph to Machine Learning). This pipeline utilizes our Positive and Unlabeled (PU) learning algorithm, PULSCAR (Positive Unlabeled Learning Selected Completely At Random), and incorporates path-based feature extraction from ProteinGraphML.

Results: KG2ML was applied to 12 diseases, including Bipolar Disorder, Coronary Artery Disease, and Parkinson's Disease, to infer disease-associated genes not explicitly recorded in DDKG. For several of these diseases, 14 out of the 15 top-ranked genes lacked prior explicit associations in the DDKG but were supported by literature and TINX (Target Importance and Novelty Explorer) evidence. Incorporating PULSCAR-imputed genes as positives enhanced XGBoost classification, demonstrating the potential of PU learning in identifying hidden gene-disease relationships.

Conclusion: The observed improvement in classification performance after the inclusion of PULSCAR-imputed genes as positive examples, along with the subject matter experts' (SME) evaluations of the top 15 imputed genes for 12 diseases, suggests that PU learning can effectively uncover disease-gene associations missing from existing knowledge graphs (KGs). By integrating KG data with ML-based inference, our KG2ML pipeline provides a scalable and interpretable framework to advance biomedical research while addressing the inherent limitations of current KGs.

Despite the growing availability of nuclear genomic data, mitochondrial genes remain the most widely used molecular markers in mammalian systematics. However, a quantitative assessment of the phylogenetic information content of mitochondrial loci compared to nuclear loci has never been carried out. Here, we apply a sparse learning approach based on Lasso regression to evaluate the contribution of alignment sites to phylogenetic likelihoods, providing the first estimates of phylogenetically effective lengths for markers commonly used in mammalian systematics. Analyzing more than 30,000 complete mammalian mitochondrial genomes and nuclear panels composed of either 100 randomly selected complete coding sequences or of partial gene segments from conventional markers, we examined phylogenetic informativeness at two taxonomic levels: within-species and among-species. On average, ∼32% of mitochondrial sites and ∼38% of nuclear sites were classified as phylogenetically informative. We found that the number of phylogenetically informative sites were positively correlated with total gene length. Therefore, longer mitochondrial genes, particularly ND5, COX1, and CYTB, harbored the largest numbers of informative sites. Although nuclear coding sequences contained, on average, more informative sites, mitochondrial genes also yielded consistent resolution of among-species relationships. Overall, our results provide the first large-scale, quantitative comparison of phylogenetic information content across mammalian mitochondrial and nuclear genes, offering a principled framework for marker selection in future systematics studies that can be broadly applied to any lineage.

Protein-protein interactions (PPIs) are vital for regulating various cellular functions and understanding how diseases are developed. The traditional ways to identify the PPIs are costly and time-consuming. In recent years, the disruptive advances in deep learning (DL) have transformed computational PPI prediction by enabling automatic feature extraction from protein sequences and structures. This survey presents a comprehensive analysis of DL-based models developed for PPI prediction, including convolutional neural networks (CNNs), recurrent neural networks (RNNs), deep neural networks (DNNs), graph convolutional networks (GCNs), and ensemble architectures. The review compares their feature representations, learning strategies, and evaluation benchmarks, emphasizing their strengths and limitations in capturing complex dependencies and structural relationships. In addition, the paper elaborates on different benchmarks and biological databases that are commonly used in different experiments for performance comparison. Finally, we outline open challenges and future research directions to enhance model generalization, interpretability, and integration with biological knowledge for reliable PPI prediction.

Background: Rapid detection of bacterial infections through leukocyte activation analysis could significantly reduce diagnostic timeframes from days to hours. Traditional methods like flow cytometry and biomarker assays face limitations including technical complexity, equipment requirements, and delayed results.

Methods: We developed a dual artificial neural network system combining stain-free light microscopy with microfluidic technology to detect morphological changes in activated leukocytes. YOLOv4 networks were trained using five-fold cross-validation on images of chemically stimulated leukocyte subpopulations (lymphocytes, monocytes, and neutrophils) and validated against flow cytometry. The system was tested on whole blood samples spiked with E. coli at clinically relevant concentrations (10-250 CFU/mL).

Results: The optimized four-class network achieved high performance metrics for lymphocytes (F1 score: 80.1% ± 2.5%) and neutrophils (F1 score: 91.7% ± 1.7%), while a specialized binary classifier for monocytes achieved 97.0% ± 2.8% F1 score. In bacteria-spiked whole blood experiments, the system successfully detected activated leukocytes within 30 min at concentrations approaching clinical blood culture detection limits (11.11 ± 4.79 CFU/mL). Neutrophils showed rapid activation peaking at 1-3 h, while lymphocyte activation increased gradually over 6-12 h, consistent with innate versus adaptive immune response kinetics.

Conclusion: This AI-assisted microscopy platform enables rapid, label-free detection of leukocyte activation in response to bacterial infection with minimal sample handling and no requirement for specialized staining or trained technicians. The technology demonstrates potential for accelerating infection diagnosis and could be extended to other pathologies with morphological manifestations.

Accurate variant calling refinement is crucial for distinguishing true genetic variants from technical artifacts in high-throughput sequencing data. While heuristic filtering and manual review are common approaches for refining variants, manual review is time-consuming, and heuristic filtering often lacks optimal solutions, especially for low-coverage data. Traditional variant calling methods often struggle with accuracy, especially in regions of low read coverage, leading to false-positive or false-negative calls. Advances in artificial intelligence, particularly deep learning, offer promising solutions for automating this refinement process. Here, we present a Transformers-based framework for genetic variant refinement that leverages self-attention to model dependencies among variant features and directly processes VCF files, enabling seamless integration with standard pipelines such as BCFTools and GATK4. Trained on 2 million variants from the GIAB (v4.2.1) sample HG003, the framework achieved 89.26% accuracy and a ROC AUC of 0.88. Across the tested samples, VariantTransformer improved baseline filtering accuracy by 4%-10%, demonstrating consistent gains over the default caller filters. When integrated into conventional variant calling pipelines, VariantTransformer outperformed traditional heuristic filters and, through refinement of existing caller outputs, approached the accuracy achieved by state-of-the-art AI-based variant callers such as DeepVariant, despite not operating as a standalone caller. By positioning this work as a flexible and generalizable framework rather than a single-use model, we highlight the underexplored potential of Transformers for variant refinement in genomics. This study contributes a blueprint for adapting Transformer architectures to a wide range of genomic quality control and filtering tasks. Code is available at: https://github.com/Omar-Abd-Elwahab/VariantTransformer.

Background: Integrating short- and long-read sequencing technologies has become a promising approach for achieving accurate and comprehensive genomic analysis. Although short-read sequencing (Illumina, etc.) offers high base accuracy and cost efficiency, it struggles with structural variant (SV) detection and complex genomic regions. In contrast, long-read sequencing (PacBio HiFi) excels in resolving large SVs and repetitive sequences but is limited by throughput, higher insertion or deletion (indel) error rates, and sequencing costs. Hybrid approaches may combine these technologies and leverage their complementary strengths and different sources of error to provide higher accuracy, more comprehensive results, and higher throughput by lowering the coverage requirement for the long reads.

Methods: This study benchmarks the DNAscope Hybrid (DS-Hybrid) pipeline, a novel integrated alignment and variant calling framework that combines short- and long-read data sequenced from the same sample. The DNAscope Hybrid pipeline is a bioinformatics pipeline that runs on generic x86 CPUs. We evaluate its performance across multiple human genome reference datasets (HG002-HG004) using the draft Q100 and Genome in a Bottle v4.2.1 benchmarks. The pipeline's ability to detect small variants [single-nucleotide polymorphisms (SNPs)/indels)], SVs, and copy-number variations (CNVs) is assessed using data from the Illumina and PacBio sequencing systems at varying read depths (5×-30×). Benchmark results are compared to those of DeepVariant.

Results: The DNAscope Hybrid pipeline significantly improves SNP and indel calling accuracy, particularly in complex genomic regions. At lower long-read depths (e.g., 5×-10×), the hybrid approach outperforms stand-alone short- or long-read pipelines at full sequencing depths (30×-35×), reducing variant calling errors by at least 50%. Additionally, the DNAscope Hybrid outperforms leading open-source tools for SV and CNV detection and enhances variant discovery in challenging genomic regions. The pipeline also demonstrates clinical utility by identifying variants in disease-associated genes. Moreover, DNAscope Hybrid is highly efficient, achieving less than 90 min runtimes at single standard instance.

Conclusion: The DNAscope Hybrid pipeline is a computationally efficient, highly accurate variant calling framework that leverages the advantages of both short- and long-read sequencing. By improving variant detection in challenging genomic regions and offering a robust solution for clinical and large-scale genomic applications, it holds significant promise for genetic disease diagnostics, population-scale studies, and personalized medicine.