Latest bioRxiv papers

Category: bioinformatics — Showing 50 items

• Systematic evaluation and benchmarking of text summarization methods for biomedical literature: From word-frequency methods to language models

  Baumgärtel, F.; Bono, E.; Fillinger, L.; Galou, L.; Keska-Izworska, K.; Walter, S.; Andorfer, P.; Kratochwill, K.; Perco, P.; Ley, M.

  Abstract

  The rapid expansion of biomedical literature demands automated summarization tools that can reliably condense research articles into concise, accurate overviews. We benchmarked 62 text summarization methods - ranging from frequency-based and TextRank extractors to modern encoder-decoder models (EDMs) and large language models (LLMs) - on a set of 1,000 biomedical abstracts for which author-generated highlights sections were available as reference summaries. Models were evaluated using a composite suite of metrics covering lexical overlap (ROUGE-1/2/L, BLEU, METEOR), embedding-based semantic similarity (RoBERTa, DeBERTa, all-mpnet-base-v2), and factual consistency (AlignScore). Our results indicate that general-purpose language models (LMs) achieve the highest overall scores across both lexical and semantic metrics, outperforming both reasoning-oriented and domain-specific models. Within the general-purpose group, medium-sized models, typically runnable on a single node, often outperform frontier-scale counterparts, suggesting an optimal balance between model capacity and computational efficiency. Statistical extractive methods lag behind all neural approaches. These findings provide a systematic reference for selecting summarization tools in biomedical research and highlight that broad pretraining remains more effective than narrow domain adaptation for generating high-quality scientific summaries.

  bioinformatics2026-07-16v4

• DIOPT: the DRSC Integrative Ortholog Prediction Tool, 2026 update

  Hu, Y.; Comjean, A.; Gao, C.; Yamamoto, S.; Mohr, S.; Perrimon, N.

  Abstract

  Mapping orthologous proteins is a critical step for cross-species literature mining, data integration, experimental design, and more, making the ability to quickly predict orthologs across species a key tool for functional genomic studies. The DRSC Integrative Ortholog Prediction Tool (DIOPT) was initially developed in 2011 to provide a centralized portal for identifying predicted orthologs among major model organisms. By integrating results from multiple ortholog prediction algorithms, DIOPT allows users to compare predictions across methods and prioritize high-confidence ortholog relationships. Over the years, we regularly updated the underlying genome annotations and refreshed predictions from each integrated algorithm. In addition, both the number of supported species and the number of ortholog prediction algorithms incorporated into the platform have grown. The web portal has also been enhanced with new features designed to improve usability, facilitate data exploration, and support a broader range of research applications. We also developed a sister version of DIOPT tailored specifically for arthropod species; this enables researchers working with a diverse set of insects and related organisms to perform ortholog mapping and comparative analyses more effectively. Together, these developments ensure that DIOPT remains a robust and broadly useful resource for functional genomics research.

  bioinformatics2026-07-16v3

• JanusX: an integrated and high-performance platform for scalable genome-wide association studies and genomic selection

  Fu, J.; Jia, A.; Wang, H.; Liu, H.-J.

  Abstract

  As genomic datasets expand in both sample size and marker density, genome-wide association studies (GWAS) and genomic selection (GS) require workflows that remain statistically rigorous, computationally efficient, and reproducible across the full analysis path, from genotype matrix to decision-relevant outputs. Here we present JanusX, an integrated high-performance framework that provides a streamlined, user-oriented workflow for GWAS and GS by unifying data handling, model execution, and visualization. Across simulated and real datasets, JanusX maintained high concordance with established baselines while substantially reducing runtime and memory usage. In GWAS, JanusX achieved up to a 19-fold speedup over GEMMA in linear mixed model (LMM) inference, and implemented additional LMM inference based on a sparse genomic relationship matrix with GRAMMAR-Gamma calibration, alleviating computational and memory bottlenecks in large-scale cohorts. JanusX also provides a FarmCPU implementation within its GWAS module, achieving a median 11.4-fold runtime improvement and reducing peak memory usage by 84.9% relative to rMVP. In GS, JanusX integrates an optimized best linear unbiased prediction (BLUP) backend that adaptively selects sample- and SNP-space solvers and incorporates a Preconditioned Conjugate Gradient (PCG) solver. This implementation efficiently completes five-fold cross-validation of 500k individuals x 500k single-nucleotide polymorphisms (SNPs) in 35.1 minutes with only 14.3 gibibyte (GiB) of peak memory. Beyond BLUP, JanusX integrates Bayesian and machine-learning predictors under a single interface with compact automatic tuning to ensure robust cross-model performance. JanusX therefore enables efficient locus discovery and genomic prediction under consistent analytical assumptions, even in large-scale cohorts.

  bioinformatics2026-07-16v2

• M6AFormer Prioritizes Unannotated Functional m6A Candidate Sites in the Human m6A Epitranscriptome

  Niu, Z.; Liu, C.; Gu, L.

  Abstract

  N6-methyladenosine (m6A) is a pervasive RNA modification with critical roles in post-transcriptional regulation, yet accurate transcriptome-wide identification of functional m6A sites remains challenging. Here, we present M6AFormer, a hybrid deep-learning framework that combines convolutional feature extraction with a lightweight Transformer to capture both local sequence motifs and broader contextual dependencies. M6AFormer consistently outperformed representative m6A predictors, including MST-M6A, CLSM6A and deepSRAMP. Transcriptome-wide scanning revealed a large repertoire of previously unannotated candidate m6A sites that retained hallmark m6A features, including canonical motif enrichment, characteristic spatial distribution and preferential overlap with m6A writer and reader binding regions. Importantly, M6AFormer-predicted sites were broadly associated with genetic and disease-relevant features, including SNPs, sequence variants and GWAS-linked loci, suggesting their potential contribution to human disease mechanisms. Finally, experimental validation confirmed a previously unreported m6A site in NEU4 mRNA and demonstrated its functional impact on cancer cell migration. Together, M6AFormer provides an accurate, interpretable and biologically informative framework for m6A site discovery.

  bioinformatics2026-07-16v2

• MSstatsBioNet: Integrating Statistical Analyses with Prior Knowledge Biomolecular Networks for Quantitative Proteomics and Phosphoproteomics

  Wu, A.; Kohler, D.; Navada, P. P.; Robbins, J. E.; Boyle, G. E.; Boshart, A.; Karis, K.; Neefjes, J.; Konvalinka, A.; Sarthy, J.; Pino, L.; Gyori, B. M.; Vitek, O.

  Abstract

  A common outcome of quantitative mass spectrometry-based proteomic and phosphoproteomic experiments is a list of proteins that are differentially abundant between conditions. However, biological interpretation requires evaluation in the context of prior knowledge of biological mechanisms and protein function. One approach to facilitate mechanistic biological interpretation is to integrate such lists with biological network databases, built from manually curated resources and text mining systems. This manuscript automates this process with MSstatsBioNet, a Bioconductor package that integrates MSstats, a family of open-source packages for detecting differentially abundant proteins, and INDRA, a system that extracts biomolecular networks from biomedical literature using text mining and merges those networks with the content of curated knowledge bases. Taking as input a list of differentially abundant proteins from MSstats, MSstatsBioNet retrieves a protein subnetwork from INDRA and overlays experimental fold changes onto the underlying subnetwork. Users can then interact with the network and overlaid data, interrogating primary literature evidence to construct granular mechanistic narratives for iterative hypothesis generation. We demonstrate the utility of this approach with three case studies, two measuring changes in protein abundance and one measuring changes in phosphorylation.

  bioinformatics2026-07-16v1

• FUSED: A Functional Representation for Joint Structural and Elemental Analysis of Protein Ligand Binding Sites

  Priyankara, T. M. S.; Ellingson, L.

  Abstract

  Ligand binding site representations are central to the analysis of protein-ligand interactions, with applications in functional characterization, binding-site comparison, and ligand recognition. Most existing approaches characterize ligand binding sites at a single distance threshold from the ligand, despite substantial variability in how such thresholds are defined and the likelihood that relevant structural and compositional information evolves across spatial scales. We propose Functional Unification of Structural and Elemental Descriptors (FUSED), a multivariate functional representation that jointly models structural and elemental compositional information of ligand binding sites as functions of distance from the ligand. Structural information is captured through covariance-based descriptors derived from the CDPA framework, while chemical composition is represented through isometric log-ratio coordinates to appropriately account for compositional geometry. Treating distance from the ligand as a continuous functional domain allows the representation to capture evolving patterns that would be lost under fixed-threshold analyses and enables data-driven identification of informative distance ranges. We evaluate FUSED on two benchmark datasets: the Extended Kahraman dataset for multiclass ligand discrimination and the TOUGH-C1 dataset for binary binding-site classification tasks. Across both datasets, the proposed framework yields compact low-dimensional representations with clear discriminatory structure and competitive predictive performance relative to established alignment-based, sequence-based, and machine learning approaches, while maintaining interpretability and low computational cost. These results suggest that functional joint modeling of structural and compositional descriptors provides an effective and flexible framework for ligand binding site analysis.

  bioinformatics2026-07-16v1

• CurateMake: an auditable workflow for multi-source ITS reference database harmonisation and phylogenetic validation

  Gardette, A.; Belda, E.; Prifti, E.; Zucker, J.-D.

  Abstract

  1. Reference databases shape the taxonomic resolution, uncertainty, and reproducibility of metabarcoding analyses. For ITS barcodes, public references are distributed across repositories with different taxonomic conventions, geographic coverage, and annotation practices, creating conflicts, missing ranks, and misannotations when databases are merged or compared. 2. We introduce CurateMake, a reproducible Snakemake workflow for ITS reference database construction, harmonisation, and validation. It integrates four public sources (UNITE, BOLD, PLANiTS, and CALeDNA) and user-supplied databases, combines Catalogue of Life name harmonisation with ITSx-based region standardisation, MSA/HMM-based alignment grouping, and SATIVA phylogenetic validation. Raw, CoL-harmonised, and SATIVA-validated annotation layers are retained throughout to compare curation effects while preserving flagged records for review. 3. We evaluated CurateMake on 3.58 million ingested sequences and controlled error-injection simulations. ITSx expanded the final harmonised database to 5.19 million barcode-resolved entries by recovering ITS1 and ITS2 sub-regions from full-length ITS records. Across the full dataset, normalised intra-cluster entropy decreased from Raw to CoL-harmonised to SATIVA-validated annotations, consistent with improved taxonomic coherence. In simulations, CurateMake achieved the highest correction rate across 1%-50% corruption and, at 15% corruption, corrected 42% +/- 1% of introduced errors, compared with 28% +/- 1% for CoL alone and 0% for SATIVA without the workflow's alignment infrastructure. 4. These results show that nomenclatural harmonisation and phylogeny-informed validation address complementary error classes, with phylogenetic validation contributing measurably only within taxon-coherent alignments in this benchmark. CurateMake therefore provides a reproducible, provenance-tracked framework for auditable ITS reference database curation in metabarcoding workflows.

  bioinformatics2026-07-16v1

• SHINE: Decoding transcriptional-metabolic microenvironments through higher-order spatial integration

  Du, B.; Wong, J. W. H.; Huang, Y.

  Abstract

  Spatial omics technologies are expanding to co-profile transcriptomics and metabolomics on the same tissue slide, providing complementary views of gene expression and biochemical activity to reveal molecular programs within native tissue microenvironments. However, integrating the transcriptome and metabolome remains technically challenging due to spatial misalignment, resolution disparity, and higher-order cross-modality interactions. Here, we present SHINE, a hypergraph-based computational framework for the joint analysis of spatial gene expression and metabolic networks derived from the co-profiling slide, focusing on representation learning and cross-modality interaction. Across multiple datasets, SHINE consistently outperformed existing methods for domain segmentation and biomarker co-localization and provided interpretable insights into metabolic-transcriptional microenvironments. Specifically, in Parkinson's disease mouse models, SHINE accurately delineates dopaminergic neuron-depleted regions and reconstructs coherent dopamine-associated axes. In human lung and breast cancers, SHINE resolves tumor-associated spatial regions and identifies spatially organized gene-metabolite programs associated with the tumor microenvironment. SHINE enables scalable spatial multi-omics integration across diverse biological systems.

  bioinformatics2026-07-16v1

• Integrating suboptimal secondary structures, AI-assisted genomic synteny, and evolutionary conservation to identify bacterial ncRNA homologs beyond sequence similarity

  Panek, J.

  Abstract

  A bioinformatic approach for genome-wide identification of homologs of bacterial non-coding RNAs (ncRNAs) integrating structural similarity, genomic synteny, and evolutionary conservation is presented. The structural similarity is detected using an algorithm for genome-wide identification of loci in genomic intergenic regions (IGRs) containing sequences capable of adopting secondary structures similar to that of the query ncRNA. The algorithm scans IGR sequences using a sliding window with a predefined step. For each window, suboptimal secondary structures are predicted and compared with the template structure to compute structural similarity scores. These scores are evaluated statistically on a genome-wide scale to infer homology of the RNAs represented by the predicted structures. Loci encoding statistically significant structures are further filtered using genomic synteny of the query ncRNAs inferred from genomic annotations. ChatGPT was used to assist in identifying literature-supported biological relationships between genes with distinct functional annotations. Syntenic loci with the structures are then examined for homologs in related species, as evolutionary conservation among related species is a common feature of ncRNAs Using this approach, we predicted novel homologs of the spot42 RNA-encoding spf gene in Glaciecola and Pseudoalteromonas genomes, and ms1 RNA genes in Frankia and Bifidobacterium genomes, where previous homology searches had failed.

  bioinformatics2026-07-16v1

• Unraveling the Comedone Switch through Single-Cell Resolution of Human Acne Lesions

  Duez, T.; Rolka, T.; Torocsik, D.; Reuter, H.; Al, B.; Gallinat, S.; Baumbach, J.; Holzscheck, N.

  Abstract

  Acne vulgaris is one of the most prevalent inflammatory skin diseases worldwide, yet the molecular events initiating comedogenesis remain poorly understood. The comedone switch hypothesis proposes that acne originates from an imbalance in lineage commitment within the junctional zone of the pilosebaceous unit, promoting infundibular differentiation at the expense of sebaceous gland maintenance. However, direct evidence from human acne tissue at single-cell resolution has been lacking. Here, we integrated single-cell transcriptomic datasets from healthy skin, non-lesional skin of acne patients, and lesional acne tissue to reconstruct the earliest stages of comedogenesis. We identified a previously uncharacterized cell population in non-lesional skin with transcriptomic features consistent with a microcomedone and mapped this population across independent datasets to reconstruct the transcriptional comedone architecture. Comedonal remodeling was characterized by enhanced keratinization and inflammatory programs. Quantitative analyses supported a shift from sebaceous toward infundibular cell fate, providing first data-driven evidence for the comedone switch hypothesis in human acne. Beyond the pilosebaceous unit, we identified broader epithelial alterations, including loss of POSTN and ERRFI1 expression in basal interfollicular epidermal keratinocytes. Together, these findings provide a cell-resolved framework for human comedogenesis and identify candidate mechanisms linking genetic susceptibility, environmental triggers, and lineage imbalance within the upper hair follicle.

  bioinformatics2026-07-16v1

• Phylogenize2: robust phylogenetic methods link genes to phenotypes across host-associated and environmental microbiomes

  Kananen, K.; Tran, N.; Bradley, P. H.

  Abstract

  In microbiome studies, associations between microbial functions and the environment are often confounded by phylogeny. While some methods explicitly account for this confounder, they require information about genome content, limiting their use in biomes where few genomes have been available. To make these methods more universally accessible, we have developed Phylogenize2, a redesigned phylogeny-aware tool for linking microbial gene families to abundance phenotypes. Phylogenize2 integrates large metagenome-assembled genome collections, including both biome-specific collections from MGnify and a broadly sampled general purpose database, GlobDB, to substantially expand species coverage, allowing its application in environments like the mouse gut and ocean. In addition, by default, Phylogenize2 uses a new robust phylogenetic testing framework that has been optimized for microbial abundance data, while also allowing the use of other comparative methods such as POMS. In an experimental mouse study, Phylogenize2 identifies that Muribaculaceae with higher abundance on a high-fat diet are enriched for proteins in the thioredoxin family, with likely roles in oxidative stress. When we apply Phylogenize2 to a polar ocean study, we find that a molybdenum-dependent PaoABC/YagTSR-like aldehyde oxidoreductase system differentiates mesopelagic from surface-dwelling Flavobacteriaceae, suggesting that aldehyde detoxification may be important for organisms that degrade marine snow. Together, these results show that Phylogenize2 expands phylogeny-aware microbiome analysis beyond the human gut and can provide insight into the genetic basis of microbiome-encoded traits in diverse environments.

  bioinformatics2026-07-16v1

• Biological Continued Pretraining Reshapes the Capability Profile of a Foundation Model Without Catastrophic Forgetting

  Wang, L.

  Abstract

  It is widely assumed that continued pretraining (CPT) on a narrow, out-of-distribution corpus such as raw biological sequence must trade away a general-purpose model's broad competence --- the "alignment tax" or catastrophic-forgetting intuition. We test this directly, without any new training, by re-analyzing three checkpoints from a single lineage of a 26B-parameter Mixture-of-Experts model (Gemma-4-26B-A4B): the instruction-tuned base, the same model after biological CPT (8.7B tokens of DNA, protein, and biomedical text), and after subsequent supervised fine-tuning (SFT). Across three independent capability axes --- general knowledge/reasoning (MMLU, ARC, HellaSwag), code generation (MBPP), and biomedical knowledge (BixBench) --- we find that biological CPT does not degrade the model; it lifts it: MMLU +13 points, MBPP pass@1 nearly doubles (0.33 to 0.63), and BixBench discrimination rises sharply (MCC 0.23 to 0.92). The single measured regression is truthfulness (TruthfulQA -8.8 points), a small and interpretable domain drift. A clean vocabulary-expansion ablation (<0.4 pt on every general metric) confirms the gains are attributable to CPT, not tokenizer changes. Crucially, subsequent SFT narrows the model back: all three axes fall to near-base levels, revealing a consistent division of labor --- CPT re-organizes and lifts the shared capability substrate; SFT cashes it out onto target tasks. We argue this reframes biological sequence not as a competitor for a foundation model's capacity but as a form of structured scientific data that reshapes its capability profile, and that CPT and SFT should be budgeted as complementary rather than substitutable stages. All checkpoints, evaluation code, and per-example outputs are public.

  bioinformatics2026-07-16v1

• Evaluating the use of non-linear models in data-driven rescoring of peptide-spectrum matches

  Nameni, A.; Declercq, A.; Gabriels, R.; Degroeve, S.; Martens, L.; Bouwmeester, R.

  Abstract

  In mass spectrometry (MS)-based proteomics, computational tools match acquired tandem MS spectra to peptides from a sequence database. Machine learning increasingly supports this task through peptide-spectrum match (PSM) rescoring, in which a classifier, typically a linear semi-supervised model, refines the initial matching score. However, Mokapot allows the user to choose among different machine learning algorithms of increasing complexity, from the default linear support vector machine (LSVM) to random forest and XGBoost. Here, we use an entrapment approach to assess the effect of this increasing complexity on PSM identification and the accuracy of the estimated false discovery rate (FDR). We show that, while more complex models increase the number of identified PSMs at a fixed FDR threshold, this gain reflects a bias towards random matches from the target proteome database rather than genuine identifications. Indeed, for the most complex model, the entrapment FDR reaches 6.3% instead of the estimated 1% decoy FDR. This bias thus yields overly optimistic FDR estimates, indicating that model complexity in PSM rescoring must be carefully balanced against this overfitting risk.

  bioinformatics2026-07-16v1

• Reference Regulatory Element-Guided Gene Expression Analysis for Mechanistic Inference of Gene Regulatory Networks

  Ren, L.; Debnath, I.; Duren, Z.

  Abstract

  Regulatory genomics faces a depth-breadth gap: deep multi-omics provides regulatory detail but is difficult to scale, whereas broad expression datasets often lack the regulatory structure needed for mechanistic Gene Regulatory Network (GRN) analysis. We developed Regulatory Elements Guided Analysis (REGA), an interpretable framework that uses reference Regulatory Element (RE) catalogs to infer transcription factor (TF)-RE-gene programs from gene expression data. Across ChIP-seq, knockdown, Hi-C, cis- and trans-eQTL benchmarks, REGA prioritized functional REs, improved RE-gene and TF-gene inference over existing baselines, including methods using more data, and recovered coherent regulatory modules. In PsychENCODE snRNA-seq, REGA identified disease-associated modules and TF activities, linked regulatory dysregulation to genetic risk, and detected cross-cell-type neuronal-glial programs. In spatial transcriptomics, REGA linked cell-intrinsic regulatory programs with intercellular ligand-receptor communication; in Perturb-seq, it mapped perturbation responses to trait-associated regulatory architectures. REGA enables scalable, interpretable GRN analysis across expression datasets.

  bioinformatics2026-07-16v1

• Differential multi-omics analysis of pulmonary arterial hypertension microvascular endothelial cells for differential drug response

  Hiort, P.; Weiss, A.; Krentz, J.; Schermuly, R. T.; Bogaard, H.-J.; Conrad, T.; Szulcek, R.; Baum, K.

  Abstract

  Pulmonary arterial hypertension (PAH) represents a heterogeneous group of disorders that involves complex molecular dysregulations, which are not fully captured by single-omics analyses. We apply our network-based multi-omics analysis framework, DrDimont, to transcriptomic, proteomic, phosphoproteomic, and kinase screening data from lung microvascular endothelial cells of PAH patients and controls. Thereby, we extend the functionality of DrDimont to incorporate kinase-kinase interactions during the construction of condition-specific multi-omics networks. Kinase interactions are inferred from phosphorylations of screened substrates that are weighted by kinase-substrate predictions. Differential interaction scores from the network-based analysis between PAH and control uncover alterations centered on kinases, in particular top hits relating to MAPK signaling, such as MAPK13, MAP2K, or upstream IRAK1, and other MAPK/MAP2K family members. Further highly differential nodes were ACADSB, GPX7, DSE (for proteins), and AIM1, LY96, CHSY3 (for mRNAs). When prioritizing drug candidates by mapping drug targets onto the differential network, we find high scores for the drug tacrolimus (FK506) and several anti-neoplastic MAPK inhibitors (e.g., selumetinib, trametinib), as well as agents acting on general proliferation via (mitochondrial) DNA transcription (e.g., epirubicin, topotecan). Integrating kinase activity screens into our explainable multi-omics network-based analyses reveals kinase-centered alterations and therapeutic hypotheses in PAH that complement single layer classical differential expression analyses.

  bioinformatics2026-07-16v1

• MiGenPro: A linked data workflow for phenotype-genotype prediction of microbial traits using machine learning.

  Loomans, M.; Suarez-Diez, M.; Schaap, P. J.; Saccenti, E.; Koehorst, J. J.

  Abstract

  The availability of microbial genomic data and the development of machine learning methods have created a unique opportunity to establish associations between genetic information and phenotypes. Here, we introduce a computational workflow for Microbial Genome Prospecting (MiGenPro) that combines phenotypic and genomic information. MiGenPro serves as a workflow for the training of machine learning models that predict microbial traits from genomes that have been annotated. Microbial genomes have been consistently annotated and features were stored in a semantic framework that is easy to query using SPARQL. The data was used to train machine learning models and successfully predicted microbial traits such as motility, Gram stain, optimal temperature range, and sporulation capabilities. To ensure robustness, a hyper parameter halving grid search was used to determine optimal parameter settings followed by a five-fold cross-validation which demonstrated consistent model performance across iterations and without overfitting. Effectiveness was further validated through comparison with existing models, showing comparable accuracy, with modest variations attributed to differences in datasets rather than methodology. Classification can be further explored using feature importance characterisation to identify biologically relevant genomic features. MiGenPro provides an easy to use interoperable workflow to build and validate models to predict phenotypes from microbes based on their annotated genome.

  bioinformatics2026-07-15v3

• Radiant DIA: A Fast, Sensitive, and Accurate Search Engine for Quantitative Proteomics

  Just, S.; Cantrell, L. S.; Nichols, A.; Wang, J.; Kis, J.; Mohtashemi, I.; Platt, T.; Farokhzad, O.; Batzoglou, S.

  Abstract

  In mass spectrometry-based proteomics, robust and efficient search engines are essential for accurate peptide and protein identification and quantification. Advances in sample preparation and instrumentation have increased the demand for highly scalable processing tools, with datasets comprising hundreds or thousands of samples in single-cell and population studies. Here we present Radiant DIA, a novel Data-Independent Acquisition search engine which achieves 4x faster processing and 10x lower cloud compute costs for large experiments while ensuring rigorous control of false discovery rate (FDR) and maintaining similar sensitivity, precision, and quantitative accuracy to widely-used tools. The Radiant DIA search engine is paired with a modular pipeline deployable on cloud and desktop environments comprising individual modules for distributed re-scoring, FDR estimation, protein inference and quantification. Unlike traditional monolithic applications, this architecture enables high-performance, cloud-scale analysis without sacrificing local usability. Together, the Radiant DIA and Fulcrum Pipeline tools enhance computational efficiency to facilitate biological discovery in large-scale proteomics, as demonstrated by analyses of real-world experiments up to thousands of MS acquisitions.

  bioinformatics2026-07-15v2

• Learning causal regulatory motifs and grammars using deep learning models and massively parallel reporter assays

  Thompson, M.; Lehner, B.

  Abstract

  A central challenge in biology is to understand, predict, and engineer the 'second genetic code': how sequence encodes gene expression. Two components of this challenge are: (1) accurate prediction (and design) of gene expression from sequence and (2) mechanistic understanding of how sequence-to-expression encoding actually works in cells. A powerful general approach to this problem is to combine large scale data generation with artificial intelligence. For example, massively parallel reporter assays (MPRAs) can quantify the expression of thousands of different sequences in pooled experiments and the resulting data can be used to train deep learning models. Unlike in the case of long-context genomic language models, where transformer-based architectures are a dominant paradigm, it remains contested whether for MPRA datasets other architectural components can lead to more useful, generalizable predictors, and whether they affect model interpretability, i.e. the ability to capture causal biological mechanisms (either inherently or when using downstream interpretability or explainability techniques, "xAI"). Ablation analyses may help elucidate important architectural components, but are almost always anecdotal, unable to describe generalizable tendencies, as they are done with a single training dataset or a few testing datasets. Here, we attempt to reconcile concerns and provide guidance for MPRA model design and xAI choice by simulating at scale 1,500 motif-based genetic architectures and evaluating the ability of different model architecture-xAI pairs to first predict an outcome given a sequence as input, and second, report involved motifs and their corresponding grammar. We find that attention-based models are efficient learners, and while we recommend their use in low-data regimes, their performance is surpassed by alternative models, like dilated CNNs, under larger sample sizes. We next show that across grammars and models, current methods for motif extraction converge toward reporting the same set of motifs, which is dominated by motifs with large effect sizes. We then perform in silico experiments across models and their discovered motifs and find that these methods accurately rank motifs based on learned effect size, but that their learned effect size is systematically miscalibrated, particularly in the presence of interactions (epistasis). Finally, we propose a novel metric for identifying motifs involved in epistasis and confirm our findings across three experimental datasets. Our work provides practical guidance for modeling and interpreting massively parallel reporter assay experiments from end to end.

  bioinformatics2026-07-15v2

• Lemonite: identification of regulatory metabolites through data-driven, interpretable integration of transcriptomics and metabolomics data

  Vandemoortele, B.; Devlies, H.; Michoel, T.; Vanhaecke, L.; Vandenbroucke, R. E.; Laukens, D.; Vermeirssen, V.

  Abstract

  Biological regulation emerges from coordinated interactions between genes, proteins, and metabolites; yet, despite their central regulatory potential, metabolites remain largely absent from genome-wide gene regulatory network inference. Current transcriptomics-metabolomics integration approaches are either limited by poor interpretability or constrained by incomplete prior knowledge, preventing the systematic identification of regulatory metabolites. Here, we present Lemonite, a data-driven and interpretable framework for integrating bulk transcriptomics and metabolomics data to uncover regulatory metabolites acting on gene modules. Lemonite extends module network inference to jointly associate transcription factors and metabolites with coexpressed gene programs, without requiring prior differential analysis or complete metabolite annotation. To contextualize predictions, we construct a comprehensive gene-metabolite knowledge graph integrating over 370,000 metabolite-gene and 2.1 million protein-protein interactions. Applied to glioblastoma and inflammatory bowel disease cohorts, Lemonite identifies over 50 functionally coherent gene modules per disease, revealing established and previously uncharacterized metabolite-gene regulatory relationships. In glioblastoma, myo-inositol and phosphatidylcholines, together with IRF6, regulate mesenchymal-like immune programs, which upon integration with single-cell transcriptomics are primarily expressed in tumor-associated macrophages and monocytes. In inflammatory bowel disease, regulatory metabolites are prioritized that change the expression of their predicted target genes in vitro. Together, Lemonite provides a principled framework to explore the genome-wide regulatory potential of the metabolome and to generate biologically interpretable, experimentally testable hypotheses from multi-omics data.

  bioinformatics2026-07-15v2

• Structural and modelling insights into the dynamic association between the transcription factor and DNA

  Jin, F.; Xu, K.

  Abstract

  The DNA recognition mechanisms by transcription factor (TF) was a significant scientific issue in the gene transcription and regulation. Multiple research technology including the experimental and modelling method has been introduced into the study of this aspect. In this article bioinformatic, protein modelling and dynamic simulation method was employed to display the overview of the dynamic binding between TF and DNA. Physical properties of positional change and freedom of atoms in addition with the volume exchange and the interaction analysis revealed the flexible binding sites of this element. The association of TF increased its stability with dynamic conformational change. The different levels of resistance to the sequential fluctuations of the residues and the nucleotides in the binding site stabilize the overall structure of the complex and initiated the open of the double helix that indicated the molecular mechanisms of the recognition and regulation of the elements.

  bioinformatics2026-07-15v2

• Reliability-weighted target prioritization in CD4+ T-cell Perturb-seq: a generalizability-theory decomposition

  Cheng, C.

  Abstract

  Genome-scale Perturb-seq screens prioritize candidate targets by the strength of a perturbation's transcriptional effect. Effect strength does not answer a prior measurement question: is the readout dependable? A large effect estimated from a single guide, a single donor, or a pseudobulk of few cells need not survive replication, and for target prioritization each false lead costs a validation experiment. We treat each perturbation effect as a measurement in a crossed Target x Guide x Donor x Condition design and apply generalizability theory (Cronbach et al., 1972; Brennan, 2001) to separate the dependable part of an effect from facet-specific idiosyncrasy. Guides and donors enter as random facets; condition enters as a fixed facet and is analyzed within its levels. For each target we report a dependability profile over the facets and a joint generalizability coefficient over the two random facets, and we re-rank targets by effect magnitude weighted by that coefficient. On the released screen (Zhu et al., 2025), removing the measurement-error floor estimated from the non-targeting controls raises the number of genes with a dependable target-signal share above .10 from 40 to 7,674. Analyzed within activation states, dependability recovers the T-cell-receptor signaling module as reliably measurable only in activated cells, without recourse to gene annotation. A design study indicates that reliability is limited by the number of guides rather than the number of donors, so a future screen should add guides. Every methodological decision was recorded and adversarially reviewed, and all results regenerate from the released summary statistics.

  bioinformatics2026-07-15v1

• Computational design of a multi-epitope vaccine against M. tuberculosis

  Buhari, A.; Okutu, P.; Oyeleke, U. A.; Sivakumar, A.; Hameed, S. A.

  Abstract

  Background: Tuberculosis remains a leading global infectious killer, with BCG offering inconsistent adult protection and rising drug-resistant strains demanding novel vaccine strategies. We report the first multi-epitope vaccine construct simultaneously targeting three previously unexplored Mycobacterium tuberculosis virulence proteins; EccB3, MycP, and polyketide synthase which collectively govern nutrient acquisition, ESX secretion integrity, and innate immune evasion. Methods: Using a reverse vaccinology pipeline, B-cell, CTL, and HTL epitopes were predicted, filtered for allergenicity, toxicity, and IFN-{gamma} induction, then assembled into an 823-residue chimeric construct incorporating beta-defensin and PADRE adjuvants with AAY/GPGPG linkers, covering ~90% global HLA diversity. The construct underwent AlphaFold structure prediction, 3DRefine refinement, disulfide engineering, PROCHECK/ProSA validation, ClusPro 2.0 docking against TLR1/TLR2, and C-IMMSIM immune simulation. Results: The construct (82.3 kDa, instability index 32.48) showed strong structural quality (94.7% favoured Ramachandran residues), stable TLR1/TLR2 binding (weighted energy: -1,371.0 kcal/mol), and robust in silico immune responses and durable memory cell formation following booster simulation. Conclusion: This computationally validated construct represents a promising multi-target TB vaccine candidate warranting experimental advancement.

  bioinformatics2026-07-15v1

• Integrative computational toxicology reveals PFOS and PFHxS associated inflammatory keratinocyte niches in psoriasis through exposure transcriptomics, single-cell spatial mapping and token-aware virtual perturbation

  Ma, J.; Yu, Q.

  Abstract

  Per- and polyfluoroalkyl substances (PFAS) are persistent toxicants with immunological, metabolic and epithelial effects, but their relevance to inflammatory skin disease remains unclear. We developed a computational toxicology framework to test whether perfluoroalkyl sulfonate programs, especially perfluorooctanesulfonic acid (PFOS) and perfluorohexanesulfonic acid (PFHxS), converge with psoriasis-associated keratinocyte inflammation. Exposure transcriptomes were derived from GSE236956, in which human embryonic stem cell-derived epithelial-lineage models were exposed to 10 M PFAS for 8-16 days. Six PFAS were prioritized using descriptors, Tanimoto similarity, toxicology evidence, adverse outcome pathway (AOP)-like key events, exposure differentially expressed gene burden and read-across support. PFAS signatures were integrated with psoriasis bulk transcriptomes, single-cell RNA sequencing, keratinocyte-state mapping, regulator and communication inference, spatial transcriptomics and token-aware Geneformer-compatible virtual perturbation. PFOS ranked highest in integrated prioritization, followed by PFHxS and perfluorooctanoic acid. PFHxS produced a smaller but directionally informative signature within a PFOS-dominant perfluoroalkyl sulfonate footprint. The shared PFOS and PFHxS program converged with psoriasis through inflammatory keratinocyte, epidermal-stress, cytoskeletal and lipid-related modules. Single-cell and spatial analyses localized the program to activated keratinocytes and inflammatory epidermal niches, with strong spatial co-localization with inflammatory keratinocyte and epidermal stress scores. Virtual perturbation prioritized S100A9, S100A8, KRT16, IL36G, CCL20, CXCL8, FABP5, KRT17, FOS, JUN and NFKBIZ as candidate effectors. These findings support an exposure-informed, experimentally testable hypothesis linking persistent perfluoroalkyl sulfonate programs to keratinocyte inflammatory niches in psoriasis.

  bioinformatics2026-07-15v1

• Reproducible-by-design: Romics Processor, a FAIR ecosystem for multi-omics and spatial-omics analysis

  Gorman, B. L.; Bhotika, H.; Jehrio, M.; Purkerson, J. M.; Carlin, F.; Nakayasu, E. S.; Misra, R. S.; Adkins, J. N.; Anderton, C. R.; Pryhuber, G.; Clair, G. C.

  Abstract

  Multi-omics and spatial-omics technologies are exploding in use, producing increasingly complex datasets. Existing bioinformatics tools are developing rapidly but fail to fully enforce the FAIR principles, leaving the field vulnerable to escalating issues in computational reproducibility. Here, we introduce a reproducible-by-design paradigm represented in an omics data processing package, RomicsProcessor. At its core, the 'Romics_object', which is a self-contained digital artifact that encapsulates the full history of the data from the original data to the fully processed state, capturing the details of the transformative steps and the required dependencies. This architecture ensures that computational workflows are fully portable and reproducible. In this manuscript, we demonstrate RomicProcessor's computational capabilities and scalability on diverse datasets, including bulk proteomics, large-scale multiplexed immunofluorescence, and multi-batch mass spectrometry imaging. Providing a robust framework for truly FAIR Data Principles-based analysis, RomicsProcessor is a blueprint for the next generation of reproducible bioinformatics tools that can dramatically accelerate discovery in multi-omics biology in the era of artificial intelligence.

  bioinformatics2026-07-15v1

• Single-cell gene networks nominate IKZF1 as an Alzheimer's microglial regulator

  Ozkurt, C.

  Abstract

  Background: Microglia drive neuroinflammation in Alzheimer's disease (AD), yet no approved therapy targets this compartment. Human genome-wide association studies consistently implicate innate immune loci in AD risk, establishing microglial transcriptional programs as therapeutically relevant but pharmacologically underexploited targets. Objective: We sought to identify transcription factors (TFs) governing microglial state transitions computationally and to nominate structurally tractable drug repurposing candidates. Methods: We applied trajectory inference (PAGA), pseudobulk DESeq2, pySCENIC gene regulatory network (GRN) inference, CellChat, and virtual screening of 1,962 approved compounds to 236,002 microglial nuclei from 84 donors (SEA-AD atlas). Results: IKZF1 was the sole target TF retained under cisTarget v10 motif constraints, with peak regulon activity in LateAD-DAM (pseudotime {rho} = +0.309) and replication in an independent bulk cohort (GSE95587; adjusted P value = .004). CellChat identified SLIT2[-&gt;]ROBO2 from multiple neuron subtypes (predominantly inhibitory interneurons) as the top predicted pathway to microglia. Tafamidis ([-&gt;]IRF8) and diflunisal ([-&gt;]PPARG) were top virtual screening hits; all evaluated compounds failed the pre-specified selectivity threshold. Conclusions: IKZF1 is prioritised as a candidate late-disease microglial TF, supported by six convergent evidence dimensions including independent bulk replication. Tafamidis and diflunisal are low-confidence repurposing hypotheses requiring experimental validation.

  bioinformatics2026-07-15v1

• A robust, sensitive phylogenetic method enables gene-level metagenomic analyses

  Tran, N.; Kananen, K.; Bradley, P. H.

  Abstract

  A key goal in the microbiome field is to move from taxonomic associations towards mechanistic hypotheses about microbial gene function. However, most methods for linking microbiome changes to specific genes are biased towards finding marker genes, with weak evidence for functional relevance. Phylogenetic regression can address this issue and has been previously applied to changes in microbial prevalence, but many environments (such as the gut in health vs. disease) are characterized more by changes in abundance, which presents unique statistical challenges. We show that when applied to real differential abundances from metagenomes, phylogenetic regression has an anti-conservative bias, indicating inflated false positives. We develop an alternative non-parametric method called "robust permutration," designed specifically for differential abundance data, and evaluate its performance against phylogenetic regression as well as several other phylogenetic comparative methods in realistic simulations of metagenomic data. These results show that robust permutration is the most powerful method that appropriately controls the false positive rate. We further apply robust permutration to a human case-control study of liver cirrhosis, revealing that Lachnospiraceae abundance in disease is linked to a previously uncharacterized iron-sulfur transcription factor encoded near homologs of the butyryl-CoA oxygen oxidoreductase system, a recently discovered system for oxygen detoxification. This illustrates how robust, sensitive phylogenetic methods can enable the generation of new molecular hypotheses directly from metagenomic case-control data.

  bioinformatics2026-07-15v1

• scDiagnostics: systematic assessment of cell type annotation in single-cell transcriptomics data

  Christidis, A.; Ghazi, A. R.; Chawla, S.; Turaga, N.; Gentleman, R.; Geistlinger, L.

  Abstract

  Although cell type annotation has become an integral part of single-cell analysis workflows, the assessment of computational annotations remains challenging. Many annotation tools transfer labels from an annotated reference dataset to a new query dataset of interest, but blindly transferring labels from one dataset to another has its own set of challenges. Often enough there is no perfect alignment between datasets, especially when transferring annotations from a healthy reference atlas for the discovery of disease states. We present scDiagnostics, a new open-source software package that facilitates the detection of complex or ambiguous annotation cases that may otherwise go unnoticed, thus addressing a critical unmet need in current single-cell analysis workflows. scDiagnostics is equipped with novel diagnostic methods that are compatible with all major cell type annotation tools. We demonstrate that scDiagnostics reliably detects complex or conflicting annotations using both carefully designed simulated datasets and diverse real-world single-cell datasets. Our evaluation demonstrates that scDiagnostics reliably identifies misleading annotations that systematically distort downstream analysis and interpretation and that would otherwise remain undetected. The scDiagnostics R package is available from Bioconductor (https://bioconductor.org/packages/scDiagnostics).

  bioinformatics2026-07-14v3

• Transplant-Agents: A Multi-Agent Artificial Intelligence Framework for Reproducibility Assessment of Post-Transplant Risk Prediction and Rejection Biomarkers

  Ding, S.; Bhattacharya, S.; Sarwal, M. M.; Sirota, M.; Butte, A.

  Abstract

  Reproducible biomarker identification and transplant rejection risk prediction remain fundamental yet unsolved challenges in transplantation medicine. Traditional approaches rely on hypothesis-driven analyses and domain expertise, limiting scalability and generalizability across diverse populations. We introduce Transplant-Agents, a data-driven multi-agent AI framework integrating large language models (LLMs) with machine learning algorithms for automated biomarker identification and rejection risk prediction. Agents interact through structured, iterative dialogue governed by predefined rules and criteria, converging on optimal biomarker sets reproducible across multiple iterations. We evaluated three multicenter clinical trial transplant datasets from ImmPort, comprising 683 patients across kidney, liver, and heart transplant cohorts. Transplant-Agents achieve AUROC scores of 0.93, 0.88, and 0.88. Feature importance analysis further confirms the stability, interpretability and potential generalizability of identified biomarkers. This work demonstrates that AI-agent frameworks can reliably reproduce established transplant biomarkers while enabling transparent, validated, and standardized risk prediction pipelines.

  bioinformatics2026-07-14v2

• Hidden sampling biases inflate performance in gene regulatory network inference

  Stock, M.; Ratajczak, F.; Bertin, P.; Hoermanseder, E.; Bengio, Y.; Hartford, J.; Falter-Braun, P.; Heinig, M.; Tong, A.; Scialdone, A.

  Abstract

  Accurate reconstruction of gene regulatory networks (GRNs) from single-cell transcriptomic data remains a major methodological challenge. Recent machine learning approaches, particularly graph neural networks and graph autoencoders, have reported improved performance, yet these gains do not consistently translate to realistic biological settings. Here, we show that a key reason for that is the way negative regulatory interactions are sampled for supervised training and evaluation. We find that widely used sampling strategies introduce node-degree biases that allow models to exploit trivial graph-structural cues rather than biological signals. Across multiple benchmarks, simple degree-based heuristics match or exceed state-of-the-art graph neural network models under these biased evaluation protocols. We further introduce a degree-aware sampling approach that eliminates these artifacts and provides more reliable assessments of GRN inference methods. Our results call for standardized, bias-aware benchmarking practices to ensure meaningful progress in supervised GRN inference from single-cell RNA-seq data.

  bioinformatics2026-07-14v2

• Taxonomic profilers and their influence on metagenomic diversity analyses

  Rondeau-Leclaire, J.; Blanchet, G.; Jacques, P.-E.; Laforest-Lapointe, I.

  Abstract

  Estimating taxonomic profiles is a central task in microbiome research. Several bioinformatic tools have been developed for this purpose, differing in algorithmic strategy, reference database flexibility, sensitivity parameters, and the type of abundance they estimate. As a result, taxonomic profiles carry an unwanted methodological signal whose driving characteristics remains understudied. While benchmarks have evaluated the performance of some of these tools, they rely on simulated data; little work has been done to compare them using real metagenomes in the presence of noise and uncharacterised diversity. Overall, the impact of taxonomic profiler choice and parameterisation on scientific conclusions remains poorly understood. Here, we leveraged 1,211 shotgun metagenomes from eight datasets to test four taxonomic profilers across 13 methodological designs. Based on diversity indices, we found substantial variability in estimated taxonomic composition depending on methodological features such as reference database and algorithmic strategy. We show that alpha diversity estimates and their associated statistical conclusions varied substantially with tool choice (particularly among k-mer-based tools) and with reference database. Beta diversity showed sensitivity to both database and parameter choices, yet this variability barely affected statistical inference. Our findings highlight the methodological sensitivity of analyses based on diversity indices and the importance for researchers to consider assessing the robustness of their results to their methodological choices. We provide a much-needed summary of tool characteristics to help researchers better understand the available bioinformatic tools and to support their methodological choices justification. This work raises awareness about the bio-informatic causes variability in diversity analyses of metagenomics data. Overall, this study underscores the importance of tool selection and parametrisation, and of conducting sensitivity analyses to support robust and reliable scientific conclusions.

  bioinformatics2026-07-14v2

• Lineage-aware stochastic modeling reveals gene-expression dynamics in development and disease

  Xing, J.; Staklinski, S. J.; Liu, Z.; Nowak, D.; Siepel, A.

  Abstract

  Gene expression changes along cell lineages, but most single-cell RNA-seq analyses treat cells as independent snapshots and ignore their phylogenetic relationships. Here we present LaVOUS, a lineage-aware probabilistic framework for modeling sparse single-cell gene-expression counts on reconstructed lineage trees. LaVOUS couples Brownian motion and Ornstein--Uhlenbeck models of latent transcriptional dynamics with negative-binomial observation models and scalable variational inference, enabling likelihood-based tests for gene-expression heritability, branch-specific expression shifts, and ancestral expression reconstruction. In simulations, LaVOUS improved detection of lineage-associated expression changes over Gaussian phylogenetic models and accurately reconstructed expression histories across expression levels. Applied to lineage-resolved single-cell datasets from metastatic lung cancer, class-switching B cells, and the developing brain, LaVOUS identified expression changes associated with metastatic progression, isotype switching, and neuronal differentiation. LaVOUS provides a general framework for studying single-cell expression dynamics across development and disease.

  bioinformatics2026-07-14v2

• Confronting spurious evaluations of computational methods in small molecule mass spectrometry

  Gupta, V.; Xu, C.; Herbst, E.; Wang, F.; Wishart, D. S.; Skinnider, M. A.

  Abstract

  Mass spectrometry-based metabolomics detects thousands of small molecule-associated signals in biological samples, but the vast majority cannot be structurally identified. Mounting interest in this metabolomic "dark matter" has spurred the development of dozens of machine-learning models for structural annotation of small molecules from their MS/MS spectra. Here, we expose a fundamental flaw in the longstanding paradigm by which these models have been evaluated. We show that a trivial machine-learning model can achieve strong performance on existing benchmarks despite entirely discarding the information contained within MS/MS spectra themselves, and without using any other auxiliary information. This performance arises because compounds with reference MS/MS spectra are structurally distinct from those found in generic chemical databases, and machine-learning models can exploit this dissimilarity by learning to predict whether a compound is likely to have been measured by MS/MS. However, we show that this confound can be overcome by using a generative model to sample decoy structures that are chemically indistinguishable from compounds in reference MS/MS libraries. The resulting benchmark cannot be solved without learning from MS/MS spectra themselves. We leverage this benchmark to compare 17 published machine-learning models for MS/MS annotation, and find that many of these models fail to outperform simple baselines and may learn little about MS/MS itself. In contrast, a subset of models show convincing evidence of generalization. Our work provides a sound foundation for developing and evaluating computational methods for small molecule MS/MS.

  bioinformatics2026-07-14v2

• DIOPT: the DRSC Integrative Ortholog Prediction Tool, 2026 update

  Hu, Y.; Comjean, A.; Gao, C.; Yamamoto, S.; Mohr, S.; Perrimon, N.

  Abstract

  Mapping orthologous proteins is a critical step for cross-species literature mining, data integration, experimental design, and more, making the ability to quickly predict orthologs across species a key tool for functional genomic studies. The DRSC Integrative Ortholog Prediction Tool (DIOPT) was initially developed in 2011 to provide a centralized portal for identifying predicted orthologs among major model organisms. By integrating results from multiple ortholog prediction algorithms, DIOPT allows users to compare predictions across methods and prioritize high-confidence ortholog relationships. Over the years, we regularly updated the underlying genome annotations and refreshed predictions from each integrated algorithm. In addition, both the number of supported species and the number of ortholog prediction algorithms incorporated into the platform have grown. The web portal has also been enhanced with new features designed to improve usability, facilitate data exploration, and support a broader range of research applications. We also developed a sister version of DIOPT tailored specifically for arthropod species; this enables researchers working with a diverse set of insects and related organisms to perform ortholog mapping and comparative analyses more effectively. Together, these developments ensure that DIOPT remains a robust and broadly useful resource for functional genomics research.

  bioinformatics2026-07-14v2

• Improved 3D Radial Phyllotaxis Trajectories for Uniform Density Distribution of Readout Directions and Sequential Binning

  Leidi, M.; Delitroz, J.; Peper, E.; Jia, Y.; Barranco, J.; Ledoux, J.-B.; Romanin, L.; Bastiaansen, J. A. M.; Schneider, J.; Franceschiello, B.

  Abstract

  Purpose: To develop 3D radial spiral phyllotaxis trajectories that provide a uniform density distribution of readout directions and support retrospec- tive sequential binning, thereby reducing ringing artifacts and improving image quality. Methods: UPhy trajectory redefines the polar angle to achieve uni- form density distribution of readout directions. FlexiPhy further decouples the azimuthal and polar ordering of interleaves through a randomized permutation, improving robustness to sequential binning. The proposed trajectories were evaluated in vivo on 10 healthy volunteers using two gradient-echo sequences on a 3T MRI scanner. Sequential temporal recon- structions were compared with reference reconstructions using structural similarity and relative L2 error metrics. Results: UPhy presents analytically demonstrated uniform density dis- tribution of readout directions. Quantitative analysis shows significantly higher SSIM values and lower relative L2 errors for FlexiPhy compared with both the original phyllotaxis and UPhy trajectories after Bonferroni correction (pcorrected < 0.05). Conclusion: FlexiPhy enables more reliable sequential binning recon- structions by reducing trajectory-induced ringing artifacts and temporal inconsistencies. Moreover, its randomized construction is not tied to a specific binning strategy, making it broadly compatible with retrospective binning approaches used in dynamic and motion-resolved MRI.

  bioinformatics2026-07-14v2

• Integrative single-nucleus multi-omics profiling identifies candidate regulators and signaling axes in Alzheimer's disease lipid-processing microglia

  Zheng, C.; Zhai, T.; Zhang, F.; Shen, L.

  Abstract

  Lipid-processing microglia are among the microglial states most strongly associated with Alzheimer's disease (AD) pathology, yet whether this association reproduces across independent cohorts, what transcriptional programs define the state, and which upstream signals and small molecules can modulate it remain unsolved. We address these questions through a cross-cohort analysis of one such substate (MG4) by integrating differential expression, transcription factor activity inference, gene set enrichment, and cell-cell communication across five independent single-nucleus RNA sequencing cohorts (n_total = 140 donors), with paired single-nucleus ATAC sequencing in one multi-omic cohort for epigenomic corroboration. A held-out cohort (n = 150 donors) supported donor-level regression of MG4 proportion on ligand expression, and two spatial transcriptomics datasets (n_total = 30 donors) related ligand expression to MG4 identity in neighboring spots. MG4 was reproducibly enriched in AD across all five cohorts (pooled log2 fold change = 0.90, p = 3.0 x 10^-4;). Expression-based inference and motif accessibility jointly nominated MITF and BACH1 as regulators of a program led by V-ATPase-driven lysosomal acidification and cholesterol efflux, a lysosomal-biogenesis signature distinct from the catabolic DAM and lipid-storage LDAM programs, with AD-specific upregulation of energy metabolism. FGF1 and TGFB2 were the most supported candidate upstream ligands, each significant in donor-level regression with further spatial evidence. Computational drug repurposing nominated ten blood-brain barrier-penetrant compounds as perturbational probes. Together, these results advance a described disease-associated microglial state into a reproducible, mechanistically framed regulatory model, providing candidate regulators, upstream ligands, and pharmacological probes for functional validation.

  bioinformatics2026-07-14v1

• MolMAE: A Surface-Centric Multimodal Masked Autoencoder for Molecular Representation Learning

  Li, J.

  Abstract

  Molecular representation learning has become a central component of modern computational drug discovery. Existing molecular foundation models mainly rely on SMILES strings, two-dimensional molecular graphs, or three-dimensional atomic coordinates. However, many molecular properties are ultimately governed by the molecular surface, where intermolecular recognition, solvation, electrostatic complementarity, and ligand-protein interactions occur. In this work, we propose MolMAE, a surface-guided multimodal masked autoencoder for molecular representation learning. MolMAE takes molecular surface point clouds, three-dimensional molecular graphs, and SMILES-derived fragment and functional-group tokens as complementary input modalities, and learns a unified multimodal molecular embedding through functional-group-aligned masked autoencoding. During pretraining, chemically corresponding local regions are jointly masked across surface, graph, fragment, and functional-group views, forcing the model to reconstruct missing geometric, physicochemical, structural, and semantic information from the remaining context. While molecular surface reconstruction serves as the primary pretraining objective, graph-, fragment-, and functional-group-level reconstruction tasks provide complementary supervision that encourages the model to capture molecular topology, bonding patterns, stereochemistry, local chemical environments, and substructure organization. In addition to reconstructing surface geometry, MolMAE reconstructs surface-associated physicochemical fields, including electrostatic potential and Fukui-related descriptors, enabling the model to learn chemically meaningful surface representations. Pretrained on approximately 261K lead-like bioactive molecules, MolMAE achieves strong performance on the ESOL benchmark under scaffold splitting and competitive performance across multiple molecular property prediction tasks. These results suggest that molecular surface-guided pretraining can complement conventional graph-, sequence-, and atom-coordinate-based molecular representations, especially for property prediction tasks influenced by exposed surface geometry and surface-associated physicochemical patterns.

  bioinformatics2026-07-14v1

• Combining transcriptomic resolutions and machine learning strategies uncovers new OXPHOS genes in Caenorhabditis elegans

  Zeballos - Goron, S.; Salinas, G.; Pazos Obregon, F.

  Abstract

  Assigning functions to genes remains a major bottleneck in biology, as many genes remain uncharacterized despite the availability of complete genome sequences. Oxidative phosphorylation (OXPHOS), the primary source of ATP in eukaryotes, exemplifies this gap. Although extensively studied in mammals, OXPHOS in other lineages has largely been inferred through sequence homology, an approach that may overlook lineage-specific components and propagate incorrect annotations. Here, we hypothesized that OXPHOS genes share characteristic transcriptional signatures that can be exploited for functional prediction. Using a curated set of 64 well-established OXPHOS genes, we combined supervised and unsupervised machine learning approaches to identify novel OXPHOS-associated genes in Caenorhabditis elegans. An ensemble of support vector machine, random forest, and k-nearest neighbors classifiers was trained on time-resolved bulk RNA-seq data using a novel informed bagging strategy and a two-round training scheme that incorporated genes annotated with limited evidence after an initial prediction round. In parallel, embryonic and adult single-cell RNA-seq datasets were used to infer co-expression networks and identify clusters enriched in known OXPHOS genes. Integrating both approaches yielded a high-confidence set of candidate genes supported by strong predictive performance on an independent test set. Several candidates lacked prior functional annotation. Functional validation of one top-ranked candidate, ril-1, showed that a ril-1 mutant displayed significantly reduced oxygen consumption, consistent with a role for ril-1 in mitochondrial respiration.Our results demonstrate that integrating complementary machine learning strategies with transcriptomic data across multiple biological resolutions enables systematic discovery of genes associated with complex cellular processes.

  bioinformatics2026-07-13v3

• SSUplex: fast, both-strand extraction and origin-sorting of small-subunit rRNA for environmental DNA metabarcoding

  O'Brien, A.; Vargas, J.; Acuna, I.; Restovic, F.; Martinez, P.; Parada, P.

  Abstract

  Ribosomal RNA metabarcoding sits at the centre of how we characterize microbial and eukaryotic communities in environmental samples, and long-read sequencing has made full-length small-subunit (SSU; 16S/18S) profiling routine. The broadly conserved primers that make rRNA such a convenient marker are also its liability: by design they co-amplify organellar (mitochondrial, chloroplast) and cross-domain SSU alongside the intended target. Left unsorted before taxonomic assignment, these passengers are systematically misclassified, and the error propagates straight into estimates of community composition and diversity. Reads must therefore be detected, extracted, and sorted by origin before they ever reach a classifier. We present SSUplex, an open-source tool that detects SSU rRNA, assigns each read to one of five origins (bacteria, archaea, eukaryota, mitochondria, chloroplast), and extracts the SSU region for downstream classification. SSUplex reimplements the extraction-and-origin logic of the widely used Metaxa2 in the Rust programming language, scans both strands, and ships as a single dependency-light binary suited to long-read (Oxford Nanopore, PacBio HiFi) and short-read data. Benchmarked against Metaxa2 on public data, SSUplex reproduces Metaxa2 origin calls on full-length reads (96.8% concordance) and matches its extraction speed on small inputs, then pulls away to run up to approximately 3.4-fold faster with approximately 35% lower peak memory at 200,000 reads, the per-sample scale a long-read amplicon run typically reaches. We characterize a genuine, measured trade-off in the origin-ranking statistic, and we identify the bacteria-versus-mitochondria boundary as the method's one intrinsically lower-confidence edge. For the now-common workflow in which origin-sorted reads are handed to a dedicated classifier rather than classified in place, SSUplex is a fast, reproducible, embeddable stand-in for Metaxa2's extraction role. Source code and a benchmark harness that regenerates every result from public data are available under the MIT license at https://github.com/ayobi/ssuplex.

  bioinformatics2026-07-13v2

• CellAwareGNN: Single-Cell Enhanced Knowledge Graph Foundation Model for Drug Indication Prediction

  Zhang, X.; Jeong, E.; Yan, C.; Feng, Y.; Lyu, L.; Guo, X.; Chen, Y.

  Abstract

  Graph foundation models have emerged as powerful tools for drug repurposing by enabling the prediction of novel drug-disease indications from large biomedical knowledge graphs. A representative example is TxGNN, which was previously developed and trained on PrimeKG, a comprehensive biomedical knowledge graph covering over 17,000 diseases. While TxGNN demonstrates strong performance, existing biomedical knowledge graphs largely lack fine-grained, cell-type-specific expression context. This limits their ability to capture disease mechanisms driven by dysregulated cellular programs, such as immune cell-specific pathways in autoimmune diseases. Moreover, prior evaluations typically test only randomly selected subsets of diseases, leaving many diseases unexamined and limiting conclusions about model performance across the full disease spectrum. To address these limitations, we first update PrimeKG to PrimeKG-U by incorporating expanded and curated biomedical knowledge and then develop TxGNN-U as a stronger graph-based baseline. Building on this foundation, we introduce CellAwareGNN, a graph foundation model that integrates single-cell genomics into PrimeKG-U. We construct a single-cell-enhanced knowledge graph, scPrimeKG, by incorporating cell-type-specific gene expression signatures from the OneK1K dataset, expanding PrimeKG from approximately 8.1 million edges and 129k nodes to over 14 million edges and 147k nodes. CellAwareGNN is pre-trained on all relation types in scPrimeKG and evaluated on drug indication prediction with explicit coverage of all diseases in the knowledge graph. CellAwareGNN consistently outperforms TxGNN and TxGNN-U. For drug indication prediction, CellAwareGNN achieves an AUPRC of 0.826, representing a 1.2% improvement over TxGNN-U (0.816) and a 3.4% improvement over TxGNN (0.799). Notably, for autoimmune diseases, CellAwareGNN attains an AUPRC of 0.864, improving by 2.0% over TxGNN-U (0.847) and 6.0% over TxGNN (0.815). Importantly, CellAwareGNN prioritizes promising repurposing candidates, including Ocrelizumab for Pemphigus via CD20-expressing B cells, Methotrexate for Pemphigus through DHFR and ATIC activity in T and B cells, and Rosiglitazone for Rheumatoid Arthritis through PPAR-{gamma} activation. These results demonstrate the value of iincorporating cell-type-specific expression context to improve both predictive performance and biological interpretability in graph-based drug repurposing.

  bioinformatics2026-07-13v2

• Optimizing automated classification for zooplankton in coastal conditions: the impact of model selection, imaging instruments, and colour information

  Hovenkamp, P. D. L.; van Walraven, L.; Ollevier, A.; van Oevelen, D.; van der Stappen, A. F.

  Abstract

  The advancement in deep learning techniques has made Convolutional Neural Networks (CNNs) a powerful tool for the fully automated classification of zooplankton images. In this study, we systematically investigate how network selection, colour information and differences in imaging instruments affect the classification of zooplankton images by comparing multiple state-of-the-art CNNs on images of zooplankton and marine snow from the in situ Continuous Particle Imaging and Classification Sensor (CPICS), Video Plankton Recorder (VPR), In Situ Ichtyoplankton Imaging System (ISIIS), and the on-board Plankton Imager (Pi-10). With differences between models of 7.8 to 19% in F1 score, we find that model selection strongly affects the classification performance, with EfficientNetV2S showing the most reliable overall performance. Moreover, differences between model architectures are largest for the least abundant classes (<100 labeled images), which implies that when these are present, careful model selection is most beneficial. The high image quality of the Pi-10 strongly increases the performance for the least abundant classes compared to the other instruments. In addition, we find a significant correlation (r = 0.597) between ImageNet the performance and F1-score on zooplankton images, which implies that more generally, a model that performs well on ImageNet will perform well for zooplankton classification. Colour information increases the F1-score of the best performing classifier with 2.8%, but provides a stronger benefit (25% F1-score) for classes with <100 images. The overall performance increase of colour information is less than expected and questions the advantage of recording colour information for zooplankton.

  bioinformatics2026-07-13v1

• Language Model Embedding Classifiers Enable Identification of Multiple Sclerosis-Associated BCRs and Repertoires

  Peet, G. C.; Owens, G. P.; Bennett, J. L.; Krishnan, A.; Macklin, W. B.

  Abstract

  Multiple sclerosis (MS) is a chronic inflammatory demyelinating disease. It affects over 2 million people worldwide but has historically been challenging to diagnose, categorize, and treat. MS has an autoimmune component that involves the production of B-cell receptors (BCRs) and immunoglobulins (Igs) that are associated with disease pathophysiology. We reanalyzed all publicly available RNA sequencing data from MS patients and extracted over 11 million BCR immunoglobulin heavy chain (IGH) sequences obtained from a variety of tissue sources. We developed a new decoder-only BCR DNA embedding model that outperforms state-of-the-art BCR and general-purpose protein language models on sequence embedding tasks. We then trained a language model classifier capable of identifying BCR sequences associated with MS. Using low-dimensional representations of whole repertoire embeddings combined with sequence-disease predictions, we can distinguish MS patient repertoires from healthy, infectious disease, or other autoimmune disease repertoires. Our models also successfully rank known MS-associated myelin-binding IgG sequences relative to controls. These findings provide a methodological foundation for BCR-based MS detection and could facilitate identification and study of disease-associated antibodies from blood.

  bioinformatics2026-07-13v1

• Biomarker Variability Limits Individualized Amyloid Time Estimation in Alzheimer Disease

  Wisch, J. K.; Jiao, Z.; Millar, P. R.; McKay, N. S.; Beric, A.; Lin, W.; Baker, B.; Stauber, J.; Preminger, S.; Jucker, M.; Barthelemy, N. R.; Chhatwal, J.; Ryan, N. S.; Schindler, S. E.; Cruchaga, C.; Benzinger, T. L. S.; Karch, C.; Bateman, R.; McDade, E.; Llibre-Guerra, J.; The Dominantly Inherited Alzheimer Network, ; The Alzheimer Disease Neuroimaging Initiative, ; Gordon, B. A.; Ances, B.; Ibanez, L.

  Abstract

  Objective: Disease progression modeling (DPM) or "amyloid time" is increasingly used to stage Alzheimer disease (AD). DPM performance depends on within-individual heterogeneity in rates of pathological accumulation as well as test-retest reliability of the biomarker. The relative contributions of these variabilities have not been systematically assessed. This would be particularly relevant if extrapolations from DPM were to be used to make individual-level predictions for research, clinical trials, or potentially future clinical practice. Methods: We conducted simulation studies incorporating empirically-derived noise properties from amyloid biomarkers to assess the contributions of inter- and intra-individual variability. Findings generalized in an autosomal dominant AD cohort with amyloid positron emission tomography (PET), cerebrospinal fluid (CSF), and plasma biomarkers and in a sporadic AD cohort with both amyloid PET and plasma biomarkers. We assessed group level DPM performance via mean average error (MAE) and root mean squared error (RMSE). At the individual level, we evaluated distinctness of distributions of biomarker levels associated with specific disease timings. Results: Inter-individual variability was the dominant source of error in temporal estimates. Intra-individual variability reduced estimate stability. Optimal performance occurred in biomarkers with positive average accumulation rates where a subset of individuals had exceptionally high levels of accumulation. In research study data, amyloid PET outperformed CSF and plasma biomarkers. Interpretation: DPM is fundamentally constrained by dynamic range, variability, and test-retest reliability of the biomarker of interest. Current DPM approaches are more robust at the group level, particularly when applied to biomarkers with more than 10-15% variability like fluid biomarkers. Funding: National Institute on Aging, Alzheimer's Association, German Center for Neurodegenerative Diseases, Raul Carrea Institute for Neurological Research, Japan Agency for Medical Research and Development, Korean Ministry of Health & Welfare and Ministry of Science and ICT, Spanish Institute of Health.

  bioinformatics2026-07-13v1

• Does ensembling improve feature attributions from sequence-to-activity models?

  Maslova, A.; Libbrecht, M.

  Abstract

  Sequence-to-activity models take as input DNA sequence and predict genomic activities such as transcription factor binding and gene expression. Applying explainable AI (xAI) methods such as DeepLIFT to these models has recently led to breakthroughs towards many genomic problems, including transcription factor binding grammar and predicting effects of genetic variants. However, there remains significant uncertainty about the reliability of sequence-to-activity interpretations.Thus, we need accurate probabilistic measures of confidence to distinguish reliable from unreliable interpretations. Towards this end, researchers have recently aimed to characterize variability acrossensembles of S2A models. However, previous work has focused on using model ensembles to improve the model predictions themselves. Here, we aim to evaluate whether model ensembles can also be used to improve feature attributions from post-hoc xAI methods. We find that ensembling attributions from multiple models improves downstream applications, including identifying transcription factor motifs and predicting regulatory genetic variants. We show that forming an ensemble using Monte Carlo Dropout (MCDropout) gets near to, but does not match, the performance of training multiple models, at much less train-time computational cost.

  bioinformatics2026-07-13v1

• Quantum Encoding Strategies for Drug Response Prediction: An Exhaustive Benchmark on a 20-Qubit Superconducting QPU

  Derouich, R.; Mathlouthi, N. E. H.

  Abstract

  We present the first systematic, hardware-executed benchmark of twelve distinct quantum data-encoding strategies for drug-response prediction on a real superconducting quantum processing unit (QPU). All experiments were conducted on the IQM Garnet 20-qubit QPU via the IQM Resonance cloud platform, using the Qrisp quantum-software framework (v 0.8.2). Each encoding was evaluated on n = 50 stratified samples drawn from the Genomics of Drug Sensitivity in Cancer dataset (GDSC2, 242 036 drug-cell-line pairs), targeting the natural-log IC50 response variable. Variational weights were optimised offline with the gradient-free COBYLA algorithm before hardware submission. Every circuit was executed with 1024 shots; the regression signal is the zero-qubit Pauli expectation value (Z0). Results show that the QAOA-inspired encoding achieves the best RMSE of 3.314 and is statistically superior (p < 0.05, Wilcoxon signed-rank test) to six of the remaining eleven encodings. Hardware-efficient entanglement structures-specifically alternating cost and mixer layers-provide a systematic advantage over purely rotational or diagonal encodings under realistic noise conditions. This work constitutes a reproducible baseline for noise-aware quantum machine learning on pharmaceutical data; all code, data, and raw QPU outputs are publicly released.

  bioinformatics2026-07-13v1

• A Bayesian Network-Based Framework for Causal Cancer Drug Target Discovery Integrating Patient and Cell Line Data

  Yoon, S. H.; Park, Y. R.; Kim, H. U.

  Abstract

  Current approaches to cancer drug target discovery face two key limitations: poor translation of cell line-derived targets to patient tumors, and the lack of causal explanation of the regulatory mechanisms underlying target prioritization. Here we present BayesTx (Bayesian Therapeutics target discovery), a Bayesian network framework that integrates patient transcriptomics data with cell line data to identify causal therapeutic targets in cancer. BayesTx projects both data domains into a shared biological space of pathway and transcription factor activities, learns domain-specific causal graphs, and merges them through weighted edge aggregation with bootstrap consensus filtering. Do-simulation on the consensus network quantifies the causal effect of each transcription factor on cancer cell viability. Applied to breast cancer using TCGA-BRCA (The Cancer Genome Atlas breast cancer cohort) and DepMap (Cancer Dependency Map) datasets, the framework ranked 47 transcription factors by predicted causal impact, with gene-level targets further derived through regulon-based propagation. Top-ranked transcription factor (TF) targets were independently supported by survival analysis in external cohort data and pharmacogenomic drug response associations. Overall, BayesTx demonstrates that cross-domain Bayesian network modeling can bridge patient and cell line data to systematically identify causal therapeutic targets in cancer.

  bioinformatics2026-07-13v1

• Learning proteomic disease trajectories with flow matching

  Hartman, E.; Karlsson, C.; Malmström, J.

  Abstract

  High-throughput proteomics has enabled detailed characterization of molecular states across health and disease. However, biological systems are inherently dynamic and methods for reconstructing continuous proteome changes remain limited. Here, we introduce proteome velocity, a framework for inferring continuous proteome trajectories from cross-sectional or sparsely sampled proteomics data using flow matching, in which a neural network learns velocity fields over proteome space. Proteome velocity estimates how rapidly and in which direction protein abundances change along a biological progression, such as disease. In mouse sepsis, covariate-conditioned velocity models resolved tissue- and pathogen-specific proteome trajectories and identified inflammatory proteins with distinct temporal activation patterns across infection routes and organ systems. In clinical COVID-19 plasma proteomes, inferred trajectories separated into distinct velocity programs associated with disease severity. These results show how generative trajectory models can transform cross-sectional or sparsely sampled proteomics into interpretable, protein-resolved representations of molecular progression.

  bioinformatics2026-07-13v1

• TOPAS: phosphoproteome data analysis and decision support platform for molecular tumor boards

  Jensen, C. B.; Sakhteman, A.; Hamood, F.; Schneider, A.; Woortman, J.; Teleanu, M.-V.; Horak, P.; Bayer, F. P.; Stange, C.; Huellein, J.; Hübschmann, D.; Henssen, A. G.; Fröhling, S.; Kuster, B.; The, M.

  Abstract

  The molecular tumor board (MTB) is central to precision oncology, providing personalized treatment recommendations based on molecular profiles of patient tumors. Genomics is instrumental for MTBs but often fails to identify clinically actionable targets, a gap that phosphoproteomics can fill. We present the tumor proteome activity status (TOPAS) platform, an end to end analysis pipeline that converts terabytes of phosphoproteomic data into patient-specific reports for MTB discussions, focusing on clinically relevant signaling linked to oncogenic mechanisms and therapeutic targets. Designed to scale with growing cohorts, the platform integrates data from 1,998 tumor samples to support patient- and cohort-level hypothesis generation. A web portal handles quality control, calculates TOPAS scores, identifies tumor antigens and immune checkpoints, and offers interactive analyses of differential protein abundance and outlier detection. The TOPAS platform is open source, addresses a critical unmet need and facilitates broader adoption of phosphoproteomics in precision oncology in the future.

  bioinformatics2026-07-13v1

• Divergent transcriptomic transition programs in Alzheimer disease: immune priming and synaptic collapse in females versus organelle membrane bifurcation in males

  Choi, M.; Kim, D.-G.; Bauermeister, S.

  Abstract

  Alzheimer's disease (AD) disproportionately affects women, who account for approximately two-thirds of prevalent cases. Despite decades of research, the mechanistic basis for this profound sex disparity remains poorly resolved. Prior transcriptomic studies have predominantly used pooled or female-enriched cohorts, obscuring whether the transition from normal cognition (NCI) to AD follows a common molecular program across sexes or reflects fundamentally divergent biology. Here, we analyzed bulk RNA-sequencing data from the dorsolateral prefrontal cortex of 624 individuals (401 females, 223 males) in the ROSMAP cohort using a transition-aware framework that separates variance instability along the disease axis from mean expression changes. We demonstrate that female and male brains exhibit structurally distinct transcriptomic transition programs. Females display a sequential, multi-tier architecture: interferon and immune variance priming (628 genes) is detectable early at the NCI-to-mild cognitive impairment (MCI) interval, which structurally precedes a massive mean-level synaptic and neuropeptide loss (8,935 genes) in AD. The female NCI-to-MCI interval alone produces 1,249 variance bifurcation events entirely absent in males. Conversely, males exhibit no early immune priming or powered mean-level changes. Instead, they collapse into a single large variance bifurcation pool (8,237 genes) heavily enriched for system-wide organelle membrane fusion, post-translational modification, and intracellular trafficking. These findings reveal that the AD transcriptomic transition is not a unitary program quantitatively modified by sex, but two distinct biological trajectories. This fundamental divergence motivates sex-stratified mechanistic models and independent biomarker development for each sex, cautioning against analytical pooling in transition-stage cohorts.

  bioinformatics2026-07-13v1

• EcoXAI: Autonomous Agentic Ecosystem for Explainable Artificial Intelligence and Biomedical Discovery

  Matsumoto, N.; Choi, H.; Freda, P. J.; Hernandez, M. E.; Wang, Z. P.; Moore, J. H.

  Abstract

  Motivation: As biomedical datasets and knowledge graphs continue to grow in size, complexity, and heterogeneity, navigating and extracting actionable insights from them presents a major bottleneck for researchers. There is a clear need for autonomous analytical solutions that can utilize recent advancements in agentic AI such as agent harnessing and loop engineering without introducing hallucination or workflow fragmentation. Researchers, regardless of technical expertise, need tools that streamline complex data analysis and deliver meaningful, actionable insights grounded in both data and established biomedical knowledge. EcoXAI addresses this by introducing a modular, customizable, containerized multi-agent system that structures analysis into explicit pipeline execution stages, lowering the computational barrier for clinical and translational researchers. Result: EcoXAI replaces monolithic AI text interfaces with an autonomous execution-driven framework with specialized bioinformatics agents for delivering proactive, data-driven insights grounded in established biological knowledge. Unlike purely LLM-driven or less integrated AI solutions prone to hallucinations or biologically implausible outcomes, EcoXAI's multi-agent framework, which leverages modern agentic management and explicit knowledge graph integration, provides greater transparency and verifiability in its reasoning. In our use case in drug repurposing for Alzheimer's Disease, EcoXAI evaluated 103 drug candidates and identified 79 novel candidates whose predictive models exceeded a randomized baseline, including the CCR5 antagonist Maraviroc, whose generated hypothesis was subsequently supported by the literature. These results demonstrate the potential of knowledge graph-grounded AI agents to accelerate hypothesis-driven biomedical research.

  bioinformatics2026-07-13v1

• amR: an R package suite to predict antimicrobial resistance in bacterial pathogens

  Ghosh, A.; Brenner, E. P.; Boyer, E. A.; McKim, A. P.; Vang, C. K.; Wolfe, E. P.; Mayer, D. A.; Lesiyon, R. L.; Ravi, J.

  Abstract

  Motivation: Identifying bacterial antimicrobial resistance (AMR) is critical for diagnostics and treatment, but resistance is a complex trait arising from myriad mechanisms spanning multiple molecular scales. Existing computational approaches often function as black boxes and rarely explore cross-species or multi-drug patterns. We developed amR, an integrated R package suite that provides a complete framework from bacterial genome data curation to interpretable AMR predictions, enabling identification of resistance mechanisms across species and drugs. Results: The amR R package suite contains three modular packages. amRdata downloads genomes and paired antimicrobial susceptibility testing data from BV-BRC and processes them, constructs pangenomes, and extracts features at gene/protein cluster, protein domain, annotated Clusters of Orthologous Groups and ResFinder AMR-associated features, and structural variant scales; data are stored in memory-efficient formats (Parquet, DuckDB). amRml trains interpretable machine learning models per species-drug combination, calculates feature importance and performance metrics, and provides rich ground for hypothesis generation and mechanism discovery. amRviz provides an interactive Shiny dashboard to explore metadata distributions and model performance across species and drugs, visualize top predictive AMR features, and analyze cross-model patterns across geographic/temporal strata. We apply the suite to Shigella sonnei, achieving a median Matthews Correlation Coefficient of 0.89 across 23 drugs and drug classes. With thousands of genomes, multi-scale features, and interpretable models, amR provides an accessible, comprehensive framework for AMR research. The amR package suite is installable via GitHub (https://github.com/JRaviLab/amR; BSD-3-Clause license).

  bioinformatics2026-07-13v1