Understanding population-level variation in food consumption and nutrient intake across age groups and geographic regions is a central problem in nutritional epidemiology and public health research. However, individual-level dietary surveys are limited by privacy concerns, reporting bias, and scalability issues. To address these limitations, we propose a computational framework that integrates aggregated food consumption data with food nutrient composition data to characterize demographic- and region-specific nutrient intake patterns, focusing on processed foods.
We combine processed-food nutrient composition data from the Ministry of Food and Drug Safety, South Korea, with aggregated processed-food consumption data from the Rural Development Administration, South Korea. Nutrient variables are harmonized through unit conversion, outlier removal, and feature scaling. Nutrient-based clustering is applied to identify food categories with coherent nutritional profiles, which are subsequently linked to consumption categories via rule-based semantic mapping. This enables the construction of population-level nutrient representations without relying on individual dietary records.
Using aggregated consumption frequencies and expenditures, we estimate age- and region-specific nutrient intake vectors as consumption-weighted averages of category-level nutrient profiles. Comparative analyses reveal differences in nutrient-based consumption structure across age groups and regions, supported by statistical tests on category distributions. Dimensionality reduction and clustering analyses in nutrient space illustrate structured relationships among food categories and population groups, highlighting systematic demographic and geographic variation in inferred nutrient intake patterns.
Overall, this study demonstrates that integrating aggregated consumption information with nutrient composition databases provides a scalable, privacy-preserving computational approach for population-level nutritional analysis and supports data-driven food system and public health research.

Histidine kinases (HKs) are part of many signaling pathways, by being implicated in two components systems (TCS). Using autophosphorylation and phosphotransfer to a response regulators (RR), they enable organisms to adapt to their environment. Most HKs are transmembrane proteins with a sensing domain outside of the cell and two catalytic domains called HisKA and HATPase. HATPase is required for interaction with the ATP and HisKA contains the phosphorylated histidine residue. HKs are involved in various environmental adaptation mechanisms, like light sensing or biochemical changes. Studying their diversity is therefore important to better understand how cells interacts with their environment. There exist incomplete HKs (iHKs) lacking either the HisKA or HATPase domain. Some iHKs with an HATPase domain possess a section of their sequence where an HisKA domain could be expected. These iHKs may contain "true" HKs, with unknown HisKA domain, that could fill gaps in various signaling pathways. In this study we analyzed 869964 sequences of iHKs having an HATPase domain but lacking an HisKA domain. We identified 18 HisKA-like profiles and did multiple meta-studies to assessed their HisKA-like characteristics. We found that their 3D structures matched the structure of known HisKA domains. We saw that the genomic context of the genes associated to these profiles contained genes implicated in signal transduction pathways. We cross-validated some of our profiles with curated annotations, as well as with a "negative dataset" made of non-HK proteins. We believe that our work could help improve the annotation of regulation pathways in prokaryotes.

Deep learning models are increasingly applied to predict bacteriophage-host infections from whole genomes, yet their decision-making processes
remain opaque. This lack of interpretability hinders model validation and limits the usefulness of their predictions for downstream biological
applications. To address this challenge, we present a reproducible Explainable Artificial Intelligence (XAI) pipeline designed to translate raw
deep learning signals into interpretable genomic insights. Our framework combines complementary attribution methods to identify important
genomic regions, which are then automatically mapped to functional annotations via a bioinformatics workflow. We illustrate the utility of this
approach across four different bacterial hosts. The results show that the reference deep learning model consistently prioritises biologically relevant
functional modules, such as tail structural proteins and lysis machinery, rather than relying on noise. Furthermore, while these high-importance
regions are predominantly unique, they functionally converge on conserved viral mechanisms. By transforming abstract importance scores into
domain-specific biological summaries, this work provides a transparent framework that facilitates interpreting and validating deep learning models
in environmental genomics.

Soil microorganisms are integral to plant invasion processes, functioning both as agents that facilitate invasion and as communities that respond to vegetation change. Despite growing evidence of microbial involvement, the effects of habitat invasibility on soil microbial communities remain poorly understood. In this study, we examined the effects of Lantana camara invasion on the diversity, composition, and functional potential of soil bacterial and fungal communities in the Corbett Tiger Reserve, India, using the Non-invaded habitat as a reference. Compared with Non-invaded habitat, Lantana camara invasion had no significant difference on bacterial alpha-diversity and fungal alpha-diversity. In contrast, the beta-diversity of both bacterial and fungal communities differed markedly between Invaded, Partially-invaded and Non-invaded habitats. Proteobacteria and Ascomycota are the most prevalent phyla in each of the three habitats in bacterial and fungal datasets respectively. Functional prediction analyses further indicated that Lantana camara invasion enhanced the representation of bacterial functions related to carbon and nitrogen cycling, including aerobic chemoheterotrophy, chitinolysis and nitrate reduction. Litter saprotrophs and ectomycorrhizal fungal guilds are significantly high in Invaded Habitat. Collectively, these findings demonstrate that Lantana invasion profoundly reshapes soil microbial communities and their functional potential.

Bioinformatic pipelines play a critical role in the analysis of amplicon sequencing data, yet their impact on downstream ecological interpretation remains insufficiently characterized. We performed a comparative analysis of three widely used pipelines—MetagenApp, QIIME2, and mothur—using a dataset of 36 pharyngeal samples (1,519,852 paired-end reads). Pipelines were evaluated in terms of read retention, computational performance, inferred feature counts, and ecological metrics. Substantial differences were observed. MetagenApp produced the highest number of inferred features (15,097), compared to mothur (10,508 OTUs) and QIIME2 (2,098 ASVs). Computational efficiency was a major differentiator: MetagenApp required only 12 min 57 s and 1.76 GB RAM, significantly outperforming QIIME2 (31 min, 14.7 GB) and mothur (2 h 02 min and 47.2 GB). Ecological analyses revealed significant differences in community structure across pipelines (PERMANOVA, R² = 0.396, p = 0.001), with consistently higher alpha diversity estimates in MetagenApp. Despite these variations, phylum-level taxonomic profiles remained broadly consistent. Our results demonstrate that pipeline selection significantly influences microbial community inference and highlight differences in computational efficiency and feature resolution across pipelines.

Pleurotus albidus is a Neotropical edible mushroom occurring from Mexico to Argentina, including Brazil's Atlantic Forest and Amazon biomes. Its bioactive Beta-glucans exhibit immunomodulatory, antioxidant, and anti-inflammatory properties, making it a promising candidate for nutraceutical and biotechnological applications. No genomic data were previously available for this species.
We report the first whole-genome sequencing and annotation of P. albidus isolate MPD161 (Oxford Nanopore MinION; ~23.6× coverage; Flye + Medaka; Galaxy Europe). The 40.54 Mb assembly (685 scaffolds; N50: 251,678 bp; GC: 49.61%) reached 98.4% BUSCO completeness (agaricales_odb12). Gene prediction with BRAKER3 and Funannotate yielded 14,015 genes, including 514 CAZymes and 1,823 secreted proteins.
antiSMASH identified 45 BGCs across six types: 1 T1PKS, 2 NRPS, 10 NRPS-like, 29 terpene, 2 terpene-precursor, and 1 NI-siderophore. Gene cluster family (GCF) analysis across nine Pleurotus genomes revealed pronounced lineage-specific biosynthetic diversification, with 10 BGCs found exclusively in P. albidus - 2 NRPS-like and 8 terpene clusters.
Genomic analysis of Beta-glucan biosynthesis identified 16 genes. Two FKS1 paralogs (GT48) form the catalytic core of the Beta-(1,3)-glucan synthase complex, activated by GTP-bound RHO1; four RHO1 paralogs and the guanine nucleotide exchange factor ROM2 govern this switch, alongside RHO2, a Beta-1,3-glucanosyltransferase (PHR2), and the stress regulator SMI1. Six GH16 proteins with the Skn1/Kre6/Sbg1 signature expand the Beta-1,6-glucan remodeling repertoire.
These findings provide a high-quality genomic reference for P. albidus and advance understanding of biosynthetic diversity and cell wall glucan biology in Basidiomycota.

Many plant-pathogenic Xanthomonas bacteria encode transcription activator-like effectors (TALEs) in their genomes. These virulence factors bind specific DNA sequences in host plants and modulate gene expression. TALE repertoires vary across strains and pathovars and differences in repeat structure affect binding preferences. Interpreting this diversity requires consistent genome-based identification and classification.

AnnoTALE is a software tool for automated TALE gene annotation and comparative analysis that introduced class-based grouping and standardized nomenclature, enabling biologically meaningful comparison of TALE repertoires across strains. Here, we present a modernization and extension of AnnoTALE focused on enhancing scalability and biological interpretability. As a prerequisite, the underlying sequence analysis library (Jstacs) was modularized and AnnoTALE separated into an independent, Maven-based project, strengthening reproducibility and maintainability. The TALE class assignment algorithm was revised and accelerated, yielding more robust and consistent classification across larger genome collections. The data model was redesigned around a relational database, enhancing data organization and access. This enables structured queries across genomes, TALE classes, and strains, improves data integrity and introduces richer, normalized strain and taxonomy metadata, increasing consistency of annotations both internally and with external reference resources.

A hosted web-based database explorer (annotale.informatik.uni-halle.de) supports biological analysis by enabling exploration of TALE class distributions across taxa, comparison of repertoires between strains, and integration of strain metadata such as geographic origin. It allows analysis of genomic context, including TALE positions within assemblies. Users can inspect individual TALE records with sequence features, annotations, and external links, supporting integrated exploratory and comparative analysis.

Pesticide-resistant insects and viruses are hard to combat with conventional chemical pesticides and significantly affect crop yield. The Colorado Potato Beetle (CPB) has a robust detoxification system and developed resistances to all major insecticide classes. Plant viruses, e.g. Cucumber Mosaic Virus (CMV), cannot be targeted by chemical pesticides, and management strategies must instead focus on eliminating transmission vectors.
RNA interference (RNAi) is an innate anti-viral response mechanism of eukaryotes that processes cytoplasmic double-stranded RNA (dsRNA) into small interfering RNAs (siRNAs) that bind to RNA-induced silencing complexes (RISC) to silence the expression of cognate target RNAs. This mechanism can be exploited by introducing dsRNA into a cell to facilitate silencing viral RNAs or vital messenger RNAs.
Advances in RNA analysis allow us to accurately identify RNA sequences that can program RISC to target specific sequences. Recent studies have been successful in protecting plants by using dsRNAs containing multiple siRNAs that can effectively silence the target RNA (Knoblich, 2025; Gago-Zachert, 2019).
The design and evaluation of these dsRNAs was labour-intensive and not standardised. We present a tool (RNAival, https://github.com/MaxiSack/RNAival) that provides a user-friendly graphical interface that enabled a standardised workflow for identifying siRNA candidates and assessing the dsRNAs processing. RNAival visualises the most potent siRNAs and interactively displays their positions on the target RNA. For dsRNA processing, it provides detailed information on the generated siRNAs, along with additional insights into potential degradation effects. Overall, RNAival offers practical support for researchers investigating siRNA-mediated RNA silencing.

The rapid expansion of viral sequence data demands taxonomic classifiers that are scalable, aligned with ICTV updates, and capable of providing interpretable evidence. We present VPF-Class 2.0, an updated successor to VPF-Class designed to meet these challenges. While retaining marker-driven protein domain detection, the new version replaces traditional rule-based voting with a lightweight supervised model based on per-genome marker-composition features.

In controlled benchmarks, VPF-Class 2.0 achieves near-perfect family-level performance and strong genus-level accuracy, significantly increasing confident annotation coverage. Under practical confidence thresholds, its performance matches or exceeds representative tools within shared taxonomic scopes. Beyond accuracy, we introduce an interpretability study that relates classification errors to the genus specificity of activated markers, offering deeper insights into viral diversity. Finally, we demonstrate the tool's applicability on large real-world viromes, showing consistent labeling and substantial agreement with graph-based classifications. This highlight presentation will showcase how VPF-Class 2.0 provides a robust, interpretable, and scalable solution for the next generation of viral metagenomics.

Microbial community profiling relies heavily on 16S amplicon sequencing given its accessibility, cost efficiency, and universal applicability across Bacteria and Archaea. However, relying on a single gene precludes insight into the genomic context and consequently functional potential. While shotgun metagenomics addresses these limitations, it remains computationally demanding and expensive. Furthermore, integrating these approaches is complicated by the frequent absence of 16S ribosomal RNA (rRNA) genes in metagenome-assembled genomes (MAGs) due to technical challenges, preventing straightforward connection of 16S rRNA sequences to their respective genomes. To address this disconnect, we screened all 3.75 million genomes in mOTUs-db, the largest systematically processed prokaryotic genome collection, recovering 1.43 million 16S sequences across 48,693 species-level units (mOTUs). Subsequently, we validated the sequence collection by assessing within-mOTU sequence heterogeneity and performed direct alignment to a curated collection of 16S data to assign robust full-length 16S representatives to mOTUs. This approach yielded representative sequences for 29,877 mOTUs, simultaneously connecting mOTUs-db to the globally distributed samples and ecological metadata in MicrobeAtlas. Notably, 7,372 of these mOTUs consist entirely of MAGs without cultivated representatives, providing direct access to the genomes corresponding to the 16S rRNA gene sequences for the first time. Applying this resource to diverse biomes, we cover an average of 92% of the human gut and 49% of the ocean community abundance. This showcases how these linkages enable extending 16S amplicon data to the underlying genomic and functional context, transforming 16S profiling into a gateway for genome-resolved microbial ecology.

Large-scale bacterial genome databases such as NCBI, AllTheBacteria, BekRep, and ProGenomes4 offer extensive genomic resources (genomes, metadata, and annotations). However, their use in ecological and applied studies remains limited due to inconsistent metadata, variable assembly quality, and data fragmentation. These limitations constrain systematic genome-level comparisons across environments, including host-associated and environmental bacteria.
EnHostDB (Environmental and Host-associated DataBase) is a curated database of ~500k high-quality bacterial genomes (80% host-associated and 20% environmental isolates). It integrates data from public resources (NCBI, ProGenomes4, IMG/JGI, CRBC) and focuses on the standardization of isolation source metadata across environmental and host-associated contexts. Genomes are quality-filtered for completeness and contamination and dereplicated using ANI to retain representative assemblies. Isolation source metadata from NCBI (~432k) and BVBRC (~68k) are curated and standardized using HiQ-BIO (High-Quality Bacterial Isolation Source Ontology). This three-level ontology (context > domain > category) maps metadata across six domains and 35 categories, enabling consistent classification across environmental (e.g., soil, freshwater) and host-associated (e.g., respiratory, gastrointestinal) habitats.
The current EnHostDB version combines genome quality metrics, taxonomy (NCBI/GTDB), and standardized metadata within a single framework for genome cohort selection. Gene-level annotations, including virulence factors, antimicrobial resistance genes, and mobile genetic elements, are under development to derive features for pathogenicity prediction models.
EnHostDB enables cross-category genomic comparisons, such as freshwater versus gastrointestinal habitats, and provides curated inputs for machine learning. The database is being developed to advance One Health research and pathogenicity prediction, with public release in the coming months.

RNA viruses such as RSV, SARS-CoV-2, DENV, and HIV have high mutation rates and compact, constrained genomes. Using large-scale public genomic datasets, we quantified context-dependent mutation rates from ancestral reconstruction on phylogenies of tens of thousands of genomes. Synonymous mutation counts were fitted with generalized linear models including sequence context, genome position, and RNA structure. We further inferred per-site fitness landscapes from the ratio of observed to neutrally expected counts.

Depending on the pathogen, the model explains up to 50% of the variance in log-counts. The clearest signature is avoidance of CpG dinucleotides, which trigger innate immunity: CpG-destroying mutations are 1.5-3-fold enriched, while CpG-creating mutations are 1.5-4-fold depleted. The strength of depletion varies across viruses, with HIV having the highest, RSV and SARS-CoV-2 intermediate, and DENV the lowest fold difference. In DENV and RSV, UpA avoidance was comparable to or stronger than CpG effects. Median counts correlate with those of their reverse complements to varying degrees across taxa and mutation types, consistent with differences in strand-symmetric mutational processes. Overall, mutational signatures reflect both host–virus interactions and features of viral replication, as well as additional yet-uncharacterized phenomena.

The log counts model also enables inference of synonymous-site fitness landscapes. These per-site landscapes highlight regions of non-coding constraint, such as the ribosome frameshift region and 3′ polypurine tract in the HIV pol gene. Mutation rates and fitness estimates inferred for HIV agree with published within-host measurements.

Most endemic viruses are seasonal and their incidence peaks during specific months of the year. Peak timing varies across regions (e.g., hemispheres), suggesting that drivers of transmission are temporally and geographically dependent. Long-term viral persistence must therefore rely on optimally timed migrations of infectious individuals into appropriate regions, with early-season introductions seeding local outbreaks. Yet, geography is often treated as neutral trait, implying that location does not affect survival. However, neglecting this dependence can lead to misinterpretation of transmission dynamics and global epidemiological data, including biased estimates of viral fitness.
We calculate how the timing and rate of introduction of infectious individuals into a region with an emerging epidemic affect viral population composition, diversity, and survival. Closed-form expressions are derived for the survival probability of an introduced infectious individual and for the distribution of the case numbers it generates at epidemic peak. Early introductions are less likely to survive than later ones, but those that do produce exponentially more infections. This trade-off creates an optimal time window around epidemic onset during which introductions account for most infections over the season. The width of this window is inversely related to the early-season growth rate of the effective reproduction number.
To validate our predictions, we compare them to estimates of the time window for relevant introductions derived from public genomic datasets of RSV-A. We find good agreement between the two, with consistent patterns across seasons except during COVID years. These results suggest that the model captures intrinsic properties of seasonal dynamics.

Chemical pollution is a major threat to biodiversity. However, among the three levels of biodiversity recognized by the Convention on Biological Diversity, genetic diversity has been received the least attention despite its fundamental role in shaping the responses of adaptive or susceptible populations to chemicals. This gap largely reflects the logistical and financial challenges in conducting large scale sequencing campaigns of wild populations alongside environmental chemical monitoring. However, the availability of relevant but disconnected data are continuously increasing, and we propose that integrating these data can characterize and help predict the level of pesticide concentrations that may alter the genetic makeup of wild populations. As a proof-of-concept, we curated and resequenced publicly available whole genome sequencing data from wild populations of six species (two invertebrates, one bird, and three mammal species) using snpArcher, an automated high-throughput variant calling pipeline. Using a landscape genomics approach, we identified single nucleotide polymorphisms (SNPs) per species whose prevalance correlates with newly modeled soil concentrations of over 100 pesticides across Europe. We assessed the strength of the associations through bootstrapping, and used the resulting set of significant SNPs per pesticide-species combination to quantify the concentration-response between pesticides and the genetic makeup of each species. From this, we derived a new, cross-species metric per pesticide, which we term genetic susceptibility distributions, to predict potential chemical impacts on the genetic makeup of wild populations. This framework shows promise for integrating genetic impacts into chemical risk assessment, with potential applications for conservation efforts and sustainable agriculture.

Compared to other ray-finned fishes, Cypriniformes is an order that is highly adapted to freshwater ecosystems. Therefore, studying it can facilitate our understanding of the adaptation process. Among all aspects of freshwater adaptation, vision is of particular interest to us because the freshwater environment is. Previous studies have investigated their visual adaptation through physiological, morphological, and genomic aspects. Owing to the advancement of sequencing and sensing technologies, a huge amount of genomic and ecological data has become available. This opens doors to uncharted dimensions. In this study, we focused on the evolution of visual opsin genes, which encode photoreceptors that sense light with different wavelengths in retinal cells. Using whole-genome sequencing data from 200 species in Cypriniformes, we predicted the visual opsin gene repertoire of the species studied. By further incorporating the visual opsin gene repertoire, spectral tuning site data, and the species tree, we inferred the gains and losses of visual opsin genes and tuning-site changes that might have occurred during the evolution of the species we studied. Our findings comprehensively reveal the visual opsin gene repertoire in currently available Cypriniformes genomes. The functional significance of these predicted changes awaits validation in future research.

Designing sustainable non-pharmaceutical strategies that balance competing objectives (i.e., hospitalization and social burden) remains a fundamental challenge in epidemic mitigation. Reinforcement learning (RL) offers a promising approach for learning intervention policies in sequential epidemic decision-making. Multi-objective RL (MORL) extends this by approximating a set of Pareto-optimal policies, enabling decision-makers to explicitly navigate trade-offs between competing objectives. To explore the problem, we use an age-structured SEIR model that was calibrated to Belgian COVID-19 incidence data and serial seroprevalence surveys to simulate epidemic spread. We employ the MOBelCov RL environment that encapsulates the compartment model and simulates different social contact restrictions to model the effects of interventions. Within this epidemic environment, we utilize the Pareto Conditioned Networks (PCN) MORL algorithm to approximate the set of Pareto-optimal policies. To learn contextually nuanced policies regarding the age-specific hospitalization risks in COVID-19, we introduce a new fairness metric, Social Burden Fairness, which weights individuals' lost social contacts by their respective risks of being hospitalized. Our experiments demonstrate how PCN learns a Pareto front of policies that dominate fixed-strategy baselines in the trade-off between hospitalizations and social burden fairness. Moreover, we note that incorporating age-based fairness regarding hospitalization risks shifts social contact reductions away from individuals with lower risks, illustrating the possibility of considering nuanced contextual information alongside disease burden for epidemic mitigation. These results demonstrate that embedding grounded fairness objectives into epidemiological models can yield actionable, sustainable, and fair mitigation policies that account for heterogeneous age-specific characteristics.