Frontiers in bioinformatics最新文献

Introduction: Peroxisome proliferator-activated receptor gamma (PPARγ) is a ligand-activated nuclear receptor with broad therapeutic relevance across various pathologies, including type 2 diabetes, obesity, cancer, and inflammatory disorders. Cannabinoids are a class of terpene-phenolic compounds from Cannabis sativa L. that have been shown to act as partial agonists of PPARγ. Among them, the acidic forms Δ⁹-tetrahydrocannabinolic acid (THCA) and cannabidiolic acid (CBDA) display higher potency than their decarboxylated counterp arts Δ⁹-tetrahydrocannabinol (THC) and cannabidiol (CBD). Despite experimental evidence supporting direct PPARγ-cannabinoid interaction, the molecular determinants governing ligand recognition within the binding pocket have not yet been comprehensively investigated.

Methods: A combination of molecular docking and molecular dynamics simulations was employed to characterize the binding modes of THC, CBD, THCA, and CBDA within the PPARγ ligand-binding domain. Docking calculations were performed on a curated set of 70 PPARγ crystal structures co-crystallized with structurally diverse ligands, exploiting thus the conformational variability of the binding pocket. The best-ranked solutions were subjected to 500 ns MD simulations and evaluated on the basis of ligand stability, persistence of polar and aromatic-aromatic interactions with the receptor, and energetic contributions estimated by MM/GBSA. Three candidate binding modes per ligand were selected and their trajectories extended to 1,000 ns.

Results: All four cannabinoids yielded at least one stable binding mode at the microsecond timescale. The cannabinoids THCA and CBDA displayed a greater number of stable binding modes than THC and CBD, a result consistent with the higher potency previously reported for these compounds in experimental studies. This behavior may be attributable to the formation of salt bridges with basic residues in the binding pocket.

Conclusion: Our findings provide a structural framework for understanding cannabinoid recognition by PPARγ. The ability of these compounds to adopt multiple binding modes may contribute to their partial agonist profile, opening new avenues for the rational design of selective PPARγ modulators with improved therapeutic properties.

Introduction: Extracellular vesicles (EVs) are increasingly recognized as active coordinators of metabolic processes rather than mere messengers. By carrying unique subsets of enzymes, metabolites, lipids, and nucleic acids, EVs can directly deliver functional metabolic machinery or dynamically alter intracellular metabolic fluxes in recipient cells. However, their role in regulating specific biochemical pathways remains largely unknown.

Methodology: In the current in silico analysis, we explored the dominant metabolic role of EV cargo using publicly available multi-omics data. For this, the top 500 mRNAs and proteins, along with miRNAs reported at least 10 times in humans across independent studies, as catalogued in the EVpedia database are considered and curated into a comprehensive dataset.

Results: Enrichment analysis of these mRNAs and proteins revealed that carbohydrate metabolic pathways, including glycolysis, the pentose phosphate pathway and the TCA cycle, were over-represented in EVs. Further, to investigate whether EVs carry miRNAs that regulate these pathways, we analyzed the miRNA targets. Enrichment analysis of EV miRNA targets mapped glycolytic regulatory genes, including HK1, HK2, PFKP and PKM. Interestingly, we also found miRNAs targeting genes encoding glucose transporters (SLC2A1, SLC2A3, SLC2A4, and SLC2A14) reported in EVs. Genomic annotation of these miRNAs revealed them to form clusters, including the miR-17-92 cluster, a well-known regulator of glycolysis.

Discussion: While the study has limitations-mainly due to the biological heterogeneity of EVs and the difficulty of standardizing cargo-the results potentially suggest that, by delivering enzymes, their mRNAs, regulatory miRNAs or combinations thereof, EVs could potentially mediate recipient cell glucose metabolism.

ClC-3 chloride channels play essential roles in cervical cancer progression by regulating lysosomal acidification, cell volume homeostasis, and chemoresistance. However, no highly selective small-molecule modulators of ClC-3 have been reported to date. Motivated by the urgent clinical need to reverse ClC-3-mediated chemoresistance and the challenge of processing massive chemical libraries under limited computational hardware resources, we propose a novel AI-driven Drug Discovery (AIDD) pipeline. Here, we present an integrated virtual drug discovery framework that transitions from traditional Computer-Aided Drug Design (CADD) by combining large-scale molecular docking, deep-learning-based rescoring, pharmacokinetic filtering, and atomistic molecular dynamics (MD) simulations. The primary advantage of this proposed scheme lies in the integration of GNINA 3D-convolutional neural network (CNN) rescoring, which significantly reduces the false-positive rates inherent in empirical scoring functions for membrane proteins. A library of ∼180,000 ZINC15 compounds was initially screened using AutoDock Vina, followed by GNINA convolutional neural network rescoring to refine predicted binding affinity and pose confidence. ADMET profiling further narrowed the candidates, providing computational proof of drug-likeness and toxicity criteria rather than experimental validation. Ultimately, only ZINC000001556308 (Lig8) satisfied all in silico criteria. To validate binding stability, we performed 100-ns all-atom MD simulations of the ClC-3-Lig8 complex embedded in a lipid bilayer. Lig8 induced reduced RMSD fluctuations, lower RMSF values across key transmembrane helices, and a more compact radius of gyration, indicating enhanced structural stabilization of ClC-3. MM/PBSA calculations confirmed favorable binding energetics dominated by van der Waals interactions, while per-residue decomposition identified PHE527, GLY283, and GLY584 as major contributors to ligand recognition. These results reveal a previously uncharacterized binding pocket within the ClC-3 transmembrane domain and highlight Lig8 as a promising lead compound for targeting ClC-3-mediated oncogenic signaling. Overall, this study establishes the first comprehensive computational framework for ClC-3 modulator discovery and provides a validated chemical scaffold for future therapeutic development against cervical cancer.

The cuprizone (CPZ) model is widely used for modeling demyelination in multiple sclerosis (MS) and for testing potential remyelination therapies. To better understand the underlying pathology of the CPZ model and evaluate its translatability, we integrated single-cell and spatial transcriptomics (ST) to investigate spatial cellular and molecular interactions during de- and remyelination in multiple brain regions. ST revealed global demyelination and neuroinflammation in the brain beyond the corpus callosum (CC), with region-specific differences. We identified oligodendroglia and microglia as two major cell types with significant transcriptomic changes in the model. CPZ-associated subclusters of oligodendroglia (marker genes Arap2, Dock10, Tenm4, Pex5l and Dock1) and microglia (marker genes ApoE, Axl, Cd9 and Lpl) were mapped to the CC by ST. During remyelination, while mature oligodendrocytes (MOL) nearly reversed their phenotype back to the control state, microglia remained associated with the demyelination phenotype. Ligand‒receptor (LR) pairing analyses predicted growth factor and phagocytic pathway enrichment during demyelination, which is consistent with changes in MS lesions, and microglia were predicted to be the major sender cells. LR pairing also predicted a high likelihood of interaction between oligodendroglia and microglia, and a novel interaction between MOL and oligodendrocyte precursor cells (OPC), underscoring their roles during de- and remyelination. Finally, astrocytes in the CPZ model had the greatest preservation of disease-associated modules in MS lesions, while MOL, OPC, and microglia showed moderate to low preservation, which overall suggests that the CPZ model has moderate translatability to chronically active MS lesions.

Introduction: Small interfering RNA (siRNA) therapeutics have extraordinary potential for targeted gene silencing. They mediate post-transcriptional gene regulation by binding to complementary messenger RNA (mRNA) sequences and degrading them, thereby preventing the production of unwanted proteins. Recent machine learning and deep learning frameworks for predicting siRNA efficacy have only achieved moderate success as these models solely rely either on handcrafted features or on sequential relations and therefore cannot capture the full complexity of siRNA-mRNA interactions.

Methods: In the above context, we propose SiaRNA, which uses a Siamese Neural Network for feature-derived representations and a bidirectional cross-attention mechanism for sequence-level relationships. It uniquely identifies mRNAs and their corresponding siRNAs as paired entities, allowing unified and context-aware modeling. Unlike previous models, which discard 2-nucleotide (2-nt) overhangs at the 3' end while using 21-nt efficacy labels, SiaRNA both trains and tests on 21-nt sequences to ensure biologically consistent predictions.

Results: Our model sets a new performance benchmark, outperforming previous state-of-the-art models. SiaRNA is trained on the HUVK dataset, achieving an accuracy of 0.881, and its generalization is confirmed by testing on the independent Simone dataset.

Discussion: The results prove SiaRNA's potential as a reliable and biologically accurate framework to guide siRNA design and improve therapeutic outcomes. On performing a case study using Patisiran siRNA and its target transthyretin (TTR) mRNA, an efficacy value of 0.7134 was observed indicating that our model can successfully identify therapeutically effective targets.

Objectives: To identify the core hypoxic injury pattern of DKD, construct a DKD risk model based on hypoxic injury-related (HIR) score, and explore the potential therapeutic targets of DKD.

Methods: DKD-related microarray-based transcriptomic analyses, single-nucleus RNA sequencing (snRNA-seq) and spatial transcriptomics were retrieved from the Gene Expression Omnibus (GEO) database. Seven HIR gene sets were obtained from various public databases. Core hypoxic genes were identified using different machine-learning algorithm. LASSO and nomogram were applied to construct a HIR risk score for cellular hypoxic damage. The detailed expression of hub gene would be showed in the single cell and kidney region. The prognostic value of the HIR score was externally validated using plasma proteomics from the United Kingdom Biobank.

Results: Five core hypoxic injury pathways in DKD were identified: Hypoxia, Autophagy, Ferroptosis, Endoplasmic Reticulum (ER) Stress, and Apoptosis. The HIR risk score was constructed based on three hub genes: CASP3, DUSP1, and ZFP36. The HIR score demonstrated high diagnostic efficiency for DKD patients. Higher HIR scores were associated with significantly infiltrated immune cells and poorer kidney function. In United Kingdom Biobank validation, the HIR score significantly improved the prediction of kidney outcomes, renal death, and secondary endpoints beyond demographic and metabolic variables (AUC increments 0.04-0.06), and correlated negatively with eGFR and positively with lipoprotein(a). The calculated tissue-level HIR scores also showed a highly significant and robust increase in the renal microenvironment of BTBR ob/ob mice.

Conclusion: These results provided a predictive model for clinical evaluation in patients with DKD and also a new insight into the role of HIR genes in the pathogenesis of DKD.

Introduction: Existing methods for gene regulatory network (GRN) inference rely primarily on gene expression data alone or on lower-resolution bulk sequencing data. Despite recent advances in integrating chromatin accessibility and RNA sequencing, inferring GRNs from paired single-cell multi-omics data remains challenging due to noise, sparsity, and complex nonlinear regulatory relationships.

Methods: We present MultiCausGRN, a graph attention network (GAT)-based framework for GRN inference from paired scRNA-seq and scATAC-seq data. The model incorporates directed prior-guided graph attention learning to capture biologically grounded regulatory directionality by integrating curated directed regulatory edges into graph representation learning. MultiCausGRN performs supervised transcription factor-target link prediction using integrated multi-omics features within a two-layer graph attention architecture.

Results: On the human PBMC multi-omics dataset, prior knowledge integration improved predictive stability and achieved a mean test AUPRC of 0.743 $\pm$ 0.049 and a mean AUROC of 0.682 $\pm$ 0.026 across five independent random seeds.

Discussion: These results demonstrate that directed prior-guided graph learning can improve the robustness and biological interpretability of GRN inference in data-limited settings. MultiCausGRN is publicly available at: https://github.com/nrr-90/MultiCausGRN.

Background: Soybean is a widely consumed oilseed crop with economic importance. It is enriched with numerous bioactive compounds that have health-promoting properties. Groundnut bud necrosis virus (GBNV) affects many agriculturally important crops, such as soybean, resulting in crop losses. GBNV is a single-stranded RNA virus from the genus Orthotospovirus and family Bunyaviridae. It is transmitted by insect vectors (thrips) and is further aggravated by secondary transmission from infected plants to others in the same field. There are no commercially available drugs against GBNV; thus, there is a need to explore phytomolecules that control the infection. This study utilizes soybean proteins (CAM, CDK1, Cul1, GSK3, HSP70, LOX, PCNA, and TCP) and GBNV proteins (CP, EGP, MP, NSP, NS, and RDRP) for the analysis of their inhibitory interactions. Physicochemical properties for both protein groups were examined. The molecular basis for the selective predictive association of these protein interactions was evaluated using in silico molecular docking approaches, dynamics simulations analysis, and post-simulation computational analyses.

Results: The results indicate that among these 50 protein-protein interactions, heat shock protein 70 (HSP70) exhibited potential inhibitory action against the RNA-dependent RNA polymerase (RDRP) with -12.12 kcal/mol binding affinity (z-score: 0.0). RDRP is an essential enzyme required for the transcription and replication of the viral genome. HSPs function in both abiotic and biotic stresses. Although further studies are required, these preliminary findings will be useful for developing new therapeutic agents against the virus, thus paving the way for researchers to find better alternatives.

Conclusion: This is the first study to combine prediction, structural validation, and interface analysis of the interaction between soybean and GBNV proteins.

Computational drug-target interaction (DTI) prediction provides a scalable alternative to costly and time-consuming experimental screening, but its reliability is limited by the scarcity of experimentally verified negative interactions. In public DTI databases, most unobserved drug-target pairs are unlabeled rather than true non-interactions. Randomly treating these unlabeled pairs as negatives can introduce label noise and reduce model reliability, particularly in cold-start scenarios involving unseen drugs or targets. To address this issue, we propose LNMGAT, a LapRLS-guided reliable pseudo-negative mining framework coupled with dual graph attention encoders. Instead of relying on experimentally confirmed negative labels or randomly sampled negatives, LNMGAT first applies Laplacian regularized least squares to drug and target similarity graphs to identify low-confidence unlabeled pairs as reliable pseudo-negatives. Drug and target representations are then learned separately on similarity-based k-nearest-neighbor graphs using graph attention networks, and their embeddings are concatenated for MLP-based interaction prediction. Across Yamanishi, Davis, KIBA, and BindingDB benchmarks, LNMGAT achieved the best AUPR in 10 of 16 evaluation settings and ranked within the top two in 14 of 16 settings. In the 12 cold-start settings, LNMGAT obtained the best AUPR in 9 cases, with absolute AUPR gains over the strongest baseline of up to 0.016 in pair cold-start prediction. External evaluation on DrugBank positive interactions and SwissDock-based molecular docking further provided database-level and in silico support for the plausibility of high-ranked predictions. Nevertheless, the biological validation in this study remains computational and database-based; no wet-lab binding assay was performed. LNMGAT therefore provides a competitive and interpretable framework for DTI prediction under negative-label uncertainty, while further experimental validation is required for its top-ranked candidates.

Introduction: Large-scale multi-site magnetic resonance imaging (MRI) improves population coverage and statistical power, but scanner- and protocol-related variability can obscure biological effects. Most harmonization methods aim to reduce site-related variance for downstream analysis, whereas less attention has been paid to where site effects are spatially expressed, whether they are reproducible across site compositions, and which acquisition parameters contribute to them.

Methods: We developed a modality-wise Linked Independent Component Analysis (LICA) framework to identify and interpret site-effect patterns in structural and resting-state functional MRI. Grey matter (GM) volume, amplitude of low-frequency fluctuation (ALFF), and regional homogeneity (ReHo) maps were analyzed separately. For each imaging measure, LICA decomposed voxel-wise maps into spatial components and subject-level loadings. Components were classified according to their associations with site labels and biological covariates, their spatial reproducibility was assessed using stepwise site-inclusion analyses, and their technical attribution was evaluated using cross-validated models based on site labels and recorded acquisition parameters. The framework was applied to ABIDE II GM maps from 913 participants across 18 sites and ALFF and ReHo maps from 795 participants across 16 sites.

Results: LICA identified site-related components across all three imaging measures. Site effects were not limited to uniform global shifts, but formed modality-specific spatial patterns. GM volume showed a dominant and highly stable whole-brain site-effect pattern, together with site-specific and regional components. In contrast, ALFF and ReHo showed more heterogeneous functional patterns, including global, focal, and scattered configurations. Site labels explained the largest proportion of loading variance, whereas recorded acquisition parameters showed modality-dependent contributions: TR and TE were more prominent for structural site effects, while FA, voxel size, TR, and scanner model contributed more strongly to functional site effects.

Discussion: The proposed framework provides a component-level diagnostic approach for multi-site MRI analysis. By mapping, stabilizing, and technically interpreting site-effect patterns, it complements conventional harmonization methods and may improve the transparency and reproducibility of multi-site structural and functional MRI studies.