ABSTRACT. Biology has recently undergone a fundamental transformation, powered by the development of sensitive molecular techniques combined with high-throughput sequencing. Major technical breakthroughs include the ability to sequence complete genomes, to finely quantify gene activity and to study its control mechanisms. Applying these techniques across species gives us a unique opportunity to study the evolution of genomic functions, and thus to better understand the mechanisms that underlie phenotypic evolution. Functional evolutionary genomics studies can bring insights into the selective pressures and molecular mechanisms that drive the emergence (or loss) of biological functions and of phenotypes. Here, we illustrate how comparative genomics approaches can bring insights into the genomic basis of convergent phenotypic evolution, in birds. The avian clade displays a spectacular diversity of phenotypes. Numerous instances of convergent phenotypic evolution are known in birds, such as the convergent loss of flight [1] or the parallel gain of vocal learning [2]. In this presentation, we will focus on one peculiar case of convergent morphological evolution : the loss of the intromittent male phallus [3]. Although an intromittent phallus was likely present in the ancestor of all amniotes, this organ was reduced or entirely lost in multiple avian lineages, including the major Neoaves clade and the Phasianidae family [3]. The evolutionary processes that led to phallus reduction or loss are still unclear, as are the genomic consequences of this major phenotypic change. Taking advantage of the availability of hundreds of avian genomic sequences, we have performed large-scale evolutionary analyses of protein-coding gene sequences and of non-coding regulatory elements, searching for genomic changes that occur in parallel with phenotypic changes. We found that hundreds of protein-coding genes and non-coding regulatory elements underwent an acceleration of their rate of evolution following this major phenotypic change. We also identify numerous gene expression differences between bird species that have retained the intromittent phallus and species that have lost this organ. While we cannot claim that these changes in expression patterns and regulatory programs are causal to the loss of the phallus, our findings illustrate the genome-wide consequences of this major phenotypic change.

References

1. Pan S, Lin Y, Liu Q, Duan J, Lin Z, Wang Y, et al. Convergent genomic signatures of flight loss in birds suggest a switch of main fuel. Nat Commun. 21 juin 2019;10(1):2756.

2. Pfenning AR, Hara E, Whitney O, Rivas MV, Wang R, Roulhac PL, et al. Convergent transcriptional specializations in the brains of humans and song-learning birds. Science. 12 déc 2014;346(6215):1256846.

3. Herrera AM, Shuster SG, Perriton CL, Cohn MJ. Developmental basis of phallus reduction during bird evolution. Curr Biol CB. 17 juin 2013;23(12):1065‑74.

ABSTRACT. Assessing ecological quality (EQ) is crucial for marine biodiversity monitoring. With the advent of High-Throughput Sequencing technologies, metabarcoding has enabled large-scale microbial community analysis through Operational Taxonomic Unit (OTU) tables, providing an alternative for EQ assessment. Machine learning (ML) models have been successfully applied for this task, but they often treat microbial abundance as the sole predictor, overlooking environmental meta-data (e.g., pH, salinity, temperature) and diversity indices (alpha and beta diversity). This study integrates metadata and diversity indices into an explainable ML framework for EQ prediction. Using SHapley Additive Explanations (SHAP), we assess the contribution of these features to model predictions across five genetic markers (V1V2, V3V4, V4, 37F, and V9). Our results highlight marker-dependent feature importance, demonstrating that while OTU-based models remain dominant, incorporating metadata improves accuracy for certain markers. This work enhances interpretability in AI-driven biomonitoring, fostering more reliable marine ecosystem assessments.

ABSTRACT. Predicting the functions of proteins remains a critical yet challenging task in computational biology. Advances in high-throughput sequencing, the expansion of protein databases, and the continuous development of artificial intelligence have led to the emergence of many computational methods dedicated to protein function prediction. In this study, we evaluated the performance of four state-of-the-art models - DeepGOPlus, DeepGraphGO, DeepGOZero, and DeepGOSE - using experimentally annotated proteins from the UniProt-KB/Swiss-Prot database. We also trained and tested these models on species-specific datasets from Arabidopsis thaliana and Oryza sativa to investigate their potential and applicability in plant protein studies. Our results showed that DeepGOPlus consistently achieved the best evaluation scores across all datasets. DeepGOSE and DeepGOZero performed comparably and only marginally outperformed DeepGraphGO in certain training attempts. Further analysis revealed that dataset stratification into training, validation, and testing sets introduced variations in Gene Ontology annotation specificity, which may have influenced model performance.

ABSTRACT. The emergence of new sequencing technologies has facilitated the acquisition of large amounts of biological data, which has proven to be a useful tool for better understanding biological systems. One way to take advantage of the potential of sequencing data is to use them to identify the relationship between biological units (e.g. genes) and phenotypical characteristics (e.g. disease outcomes). This question, formulated as a variable selection problem, remains difficult because of the size of the data and their correlation structure. To address these challenges, we studied the applicability of the knockoff (KO) procedure focusing on transcriptomic data in a classification setting. Introduced by Candès et al. in 2015, the KO variable selection procedure has shown promising results on real biological data. This method seeks to identify the truly important predictors by overcoming the correlation structure between variables while controlling the false discovery rate even in high dimensional settings. We conducted an extensive simulation study using real data to evaluate the relevance of recent methods in the context of high-dimensional classification. We also analyzed the benefits of a KO aggregation scheme to mitigate the effect of stochasticity, which is intrinsic to the KO procedure. In addition, we studied the stability of the KO framework as a measure of the reliability of variable selection. Finally, we applied the KO framework to real transcriptomic data.

ABSTRACT. Microproteins—defined as proteins comprising fewer than 100 amino acids—have long been overlooked due to challenges in detection arising from their small size and low expression. However, emerging evidence reveals that they are key regulators of translation, development, metabolism, and cellular stress responses, and are implicated in diseases such as cancer and cardiovascular disorders . Recent advances in ribosome profiling and mass spectrometry have uncovered a vast transient microproteome originating from non-canonical open reading frames (ORFs) overlapping annotated genes or located in UTRs or intergenic regions. Although many of these sequences lack signs of selection and are evolutionarily transient, some impact fitness and may represent early stages of de novo gene formation.

In this study, we introduce a novel computational framework that employs deep learning–based protein language models (pLMs) to infer the structural properties of microproteins without relying on evolutionary information. Using embeddings from ProtT5-XL, we analyzed thousands of microproteins from annotated sources (e.g., Uniprot) and potential iORF-encoded microproteins across diverse eukaryotic genomes with varying GC content. Using simple dimensionality reduction of these embeddings, we constructed a comprehensive map of the microprotein structural landscape, confirming that amino acid composition and residue ordering are the primary determinants of structure.

Then we fine-tuned a classifier that demonstrated robust performance in predicting structural categories, capturing additional signals beyond those revealed by embedding dimensionality reduction. Our results indicate that annotated microproteins occupy narrow, well-defined regions of the structural space, whereas iORF-encoded microproteins exhibit a broader, GC-dependent distribution. Specifically, low-GC iORFs are biased toward encoding transmembrane peptides, while high-GC iORFs predominantly yield disordered proteins. Moreover, certain iORF sequences fall into a “void” region not populated by canonical proteins, suggesting a region that might have been counter-selected by evolution.

Taken together, these findings challenge prior knowledge of protein-coding potential and offer fresh insights into the potential evolutionary emergence and structural diversity of microproteins. Our results characterize the structural landscapes of two distinct yet interconnected microproteomes—the annotated coding microproteome and the unannotated, noncoding counterpart—and lay the groundwork for new hypotheses about the molecular evolution of coding sequences from noncoding origins.

ABSTRACT. Microbial ecosystems constitute complex yet information-rich environments whose characterization is crucial for understanding host health and disease. Among them, the human gut microbiome has emerged as a key ”super-integrator”, owing to its dense interactions with host physiology and its established associations with a wide spectrum of pathologies. Driven by advances in high-throughput sequencing technologies and the continuous decline in associated costs, metagenomic studies have expanded exponentially, generating massive amounts of sequencing data and opening new avenues for data-driven disease. modeling.

Conventional approaches to microbiome analysis predominantly rely on the alignment of DNA sequencing reads against reference databases to infer microbial composition at the species level. While effective, these methods are inherently constrained by reference bias and limited taxonomic resolution.

Recent advances in artificial intelligence—particularly in Natural Language Processing (NLP)—offer new methodological perspectives for metagenomic data representation. In this study, we present MetagenBERT, a Transformer-based framework that relies on the foundational models DNABERT-2 and DNABERT-S for the embedding of DNA sequencing reads. Our approach encodes gut microbiome metagenome in a species-agnostic manner, enabling direct downstream application to disease classification tasks. We show that MetagenBERT attains similar performance to state-of-the-art abundance-based models for cirrhosis prediction and surpasses them in the more challenging context of type 2 diabetes detection. Furthermore, we introduce an alternative representation of metagenomic profiles based on read-level embeddings aggregated into abundance vectors, demonstrating their complementarity with conventional species-level abundance metrics.

ABSTRACT. SpecPeptidOMS directly aligns peptide fragmentation spectra to whole and undigested protein sequences. The algorithm was specifically and initially designed for peptidomics, where the aim is to identify peptides that do not result from the hydrolysis of a known protein and therefore, whose termini cannot be predicted. Thus, SpecPeptidOMS can perform alignments starting and ending anywhere in the protein sequence. The underlying computational method of SpecPeptidOMS, which is based on a dynamic programming approach, was drastically optimized. As a result, SpecPeptidOMS can process around 12,000 spectra per hour on an ordinary laptop, with alignment performed against the entire human proteome. The performance of SpecPeptidOMS was first evaluated on a publicly available data set of (nontryptic) synthetic mass spectra. Accuracy was estimated by considering the results obtained by MaxQuant on the same data set as the “ground truth”. A second series of tests on a larger, well-known proteomics data set (HEK293) highlighted SpecPeptidOMS’ additional ability to search for open modifications, a feature of interest in peptidomics but also more broadly in conventional proteomics. SpecPeptidOMS is open-source, cross-platform (written in Java), and freely available.

ABSTRACT. The characterization of centromeric DNA has recently seen a major breakthrough thanks to advances in sequencing technologies, reaching complete resolution for several human cell lines and other primates through Telomere-to-Telomere (T2T) assemblies. New bioinformatic tools have been developed to describe the main DNA type composing the centromere in most Primates, alpha satellite (AS) DNA. However, both the tools and the resulting annotations often rely on the human genome to interpret the evolutionary dynamics of other species. One example is the use of human AS families to annotate AS sequences in other species, despite the rapid evolution of these sequences. Moreover, the current model of alpha satellite evolution, which is primarily based on observations in humans, needs to be tested against data from other primates. As T2T sequencing remains costly and technically demanding, we propose an alternative approach that directly leverages long-read data. In this study, we identified AS-containing reads from long-read sequencing of a Cercopithecini species, Cercopithecus cephus. Through a de novo annotation of these sequences, we identified two families that we previously detected in two other Cercopithecini species, as well as a new family, C3, which is the least abundant but also the most ancient. These three families also appear to be spatially segmented across the genome, corresponding to distinct homogeneous evolutionary layers. To investigate the organization of alpha satellite monomers within these layers, we developed HORfinder, a tool designed to detect higher-order repeat (HOR) structures without relying on predefined family classifications. Unlike humans, C. cephus exhibits a predominantly monomeric organization of its AS, with only 0.4% of sequences forming HORs. Interestingly, these HORs are mainly found in the oldest evolutionary layers, suggesting a potential transition from HOR to monomeric organization. These findings support the idea that HOR organization is not a unique or highly specialized structure and could arise independently in multiple clades.

ABSTRACT. Understanding the genetic basis of phenotypic adaptations poses a significant challenge in evolutionary genomics. Despite the morphological and physiological diversity in mammalian traits, their coding genomes exhibit a high degree of conservation, implying that changes in gene expression and regulation are pivotal in driving phenotype evolution This study aims to identify shifts in gene expression and cis-regulatory activity and their potential role in phenotypic adaptation. Using African mole-rats as a model, renowned for their unique phenotypic adaptation traits like cancer resistance and hypoxia tolerance, we aimed to elucidate the genome-wide gene expression patterns underlying these traits that have been mainly characterised at the level of candidate genes and in individual species. Profiling gene expression in heart and liver tissues across two mole-rat species and two rodent outgroups, we used a phylogenetic comparative approach to identify genes with expression shifts within the mole-rat clade and in specific genera. These shifted genes are associated with functions pertinent to known adaptations in naked mole-rats, such as cellular respiration and glycolysis in the heart. Furthermore, our analysis revealed concordant changes in the regulatory landscape of these genes. By employing a phylogenetic comparative approach, we offer new insights into the interplay between gene expression, regulation, and phenotypic evolution in mammals. Our findings shed light on the molecular mechanisms driving the evolution of unique traits in mole-rats and potentially other mammalian species.

ABSTRACT. Inverted repeats (IRs) are genetic elements with a DNA motif (left arm) followed by a gap, or spacer, and its reverse complement (right arm) – e.g., ATACGGnnnCCGTAT. They play a key role in manybiological functions, including gene regulation, DNA replication, and genome plasticity. In this study, we aim to systematically investigate the distribution of short IRs (with gap lengths up to 20 bp) in all completely sequenced prokaryotic species by confronting observed statistics to expected ones computed by permuting DNA sequences under specific constraints. Through this systematic approach, we reveal complex patterns of IR biases with five main observations: (i) a systematic enrichment of IRs with arm lengths longer than 6 bp, (ii) a systematic depletion of palindromes (IRs with a zero-length gap) shorter than 6 bp, (iii) qualitatively different biases between coding and non-coding regions, (iv) in non-coding regions, the most frequent enrichment over bacterial species occurs for gap length of 4 bp similar to the most common loop size of RNA hairpins associated with known transcription terminators in model organisms, and (v) biases in coding regions that strongly depend on the species considered. Altogether, these findings — both corroborating and further deepening previous analyses — highlight universal evolutionary constraints as well as species-specific selective pressures that act on genome sequences, particularly on IRs.

ABSTRACT. Background: Interpretability is a topical question in recommender systems, especially in healthcare applications. An interpretable classifier quantifies the importance of each input feature for the predicted item-user association in a non-ambiguous fashion. Results: We introduce the novel Joint Embedding Learning-classifier for improved Interpretability (JELI). By combining the training of a structured collaborative-filtering classifier and an embedding learning task, JELI predicts new user-item associations based on jointly learned item and user embeddings while providing feature-wise importance scores. Therefore, JELI flexibly allows the introduction of priors on the connections between users, items, and features. In particular, JELI simultaneously (a) learns feature, item, and user embeddings; (b) predicts new item-user associations; (c) provides importance scores for each feature. Moreover, JELI instantiates a generic approach to training recommender systems by encoding generic graph-regularization constraints. Conclusions: First, we show that the joint training approach yields a gain in the predictive power of the downstream classifier. Second, JELI can recover feature-association dependencies. Finally, JELI induces a restriction in the number of parameters compared to baselines in synthetic and drug-repurposing data sets.

ABSTRACT. Protein embedding consist of producing a mathematical representation of a protein based on data such as sequence or structure. Protein embedding is widely use thanks to the last advances in artificial intelligence that allow embedding to resume proteins information (such as function, biochemical properties, ...). One of the setback to protein embedding is the dimensionality of the latest, often very large, leading to difficulties to manipulate them efficiently (the so called Curse of Dimensionality). We present here our work on protein embedding based on protein domains architecture. A protein having less domains than amino acids we hope to produced embedding of lower dimensionality that would be easier to use. We trained two models based on the Bert architecture on different training datasets using the mask language modeling objective. Our training datasets were obtained by annotating Uniprot (Trembl + SwissProt) and BFD proteins using PFAM domains and Low Complexity Region. Our models show good performances on some training sets and seems to be able to learn a good protein representation from their domains architecture.

ABSTRACT. In recent years, Knowledge Graphs (KGs) have gained significant attention for their ability to organize complex biomedical knowledge into entities and relationships. Knowledge Graph Embedding (KGE) models facilitate efficient exploration of KGs by learning compact data representations. These models are increasingly applied to biomedical KGs for link prediction, for instance to uncover new therapeutic uses for existing drugs. While numerous KGE models have been developed and benchmarked for link prediction, existing evaluations often overlook the critical issue of data leakage. Data leakage leads the model to learn patterns it would not encounter when deployed in real-world settings, artificially inflating performance metrics and compromising the overall validity of benchmark results. In machine learning, data leakage can arise when (1) there is inadequate separation between training and test sets, (2) the model leverages illegitimate features, or (3) the test set does not accurately reflect real-world inference scenarios. In this study, we implement a systematic procedure to control train-test separation for KGE-based link prediction and demonstrate its impact on models' performance. In addition, through permutation experiments, we investigate the potential use of node degree as an illegitimate predictive feature, finding no evidence of such leveraging. Finally, by evaluating KGE models on a curated dataset of rare disease drug indications, we demonstrate that performance metrics achieved on real-world drug repurposing tasks are substantially worse than those obtained on drug-disease indications sampled from the KG.

ABSTRACT. Microbial translocation occurs when bacteria migrate from the gut to the blood due to gut barrier alterations, potentially triggering persistent immune activation and affecting immune responses. Translocation can be studied using whole blood RNA sequencing techniques to analyze the “meta- transcriptome”: all the RNA from bacteria, viruses and fungi from the blood. However, a key challenge is reusing cohort data, which primarily consists of human sequences, to characterize the non-human meta-transcriptome. Cohort studies typically focus on human data while minimizing non- human contamination. From a translocation perspective, these non-human sequences become the focus, requiring human sequences to be filtered out, leaving a small fraction (~2%) to be analyzed. In previous work, Douek et al. used a sequence assembly-based pipeline for translocation analysis. We compared this approach with an assembly-free method and found that integrating both strategies into a “hybrid” pipeline improved classification performance in simulations. While real-data validation remains challenging, our results suggest this hybrid strategy enhances microbial translocation analysis.

ABSTRACT. Bioinformatics is increasingly used in various scientific works. Large amounts of heterogeneous data are being generated by scientific teams or laboratories for research purposes. Sequencing and other biological data are difficult to interpret and analyze effectively without dedicated and adapted tools. Several software tools have been developed to facilitate handling and analyses of these types of data. The Galaxy project web platform is one of these software tools that allow free access to users and facilitates the use of thousands of bioinformatics tools. Other software tools like Bioconda or Jupyter Notebook make it easier to install tools and their dependencies for bioinformatics scripts or to offer a user-friendly web interface. In addition to these tools, we can mention RStudio which is an integrated development environment (IDE) facilitating the use of R scripts. The aim of this study is to provide some guides (or helpers) to the scientific community to perform some bioinformatics or biostatistics analyses in a simpler manner. We also try with this work to democratize well-documented software tools to make them suitable for both bioinformaticians and non-bioinformaticians. We believe that user-friendly guides and real-life/concrete examples will provide end users with suitable and easy-to-use methods for their bioinformatics analysis needs. Furthermore, tutorials and examples of use will be available on our dedicated GitHub repository (https://github.com/karubiotools/AnssBin). These tutorials/examples (in English and/or French) could be used as pedagogical tools promoting bioinformatics analyses and potential answers to some bioinformatics needs. Platforms and/or services play an important role in helping scientists with their bioinformatics data analysis work. These facilities are the cornerstone of bioinformatics capacity building in the overseas islands and support the growth of nascent networks such as KaruBioNet.

ABSTRACT. Modeling microbial interactions as sparse and reproducible networks is a major challenge in microbial ecology. Direct interactions between the microbial species of a biome can help to understand the mechanisms through which microbial communities influence the system. Most state-of-the art methods reconstruct networks from abundance data using Gaussian Graphical Models, for which several statistically grounded and computationally efficient inference approaches are available. However, the multiplicity of existing methods, when applied to the same dataset, generates very different networks. In this article, we present OneNet, a consensus network inference method that combines seven methods based on stability selection. This resampling procedure is used to tune a regularization parameter by computing how often edges are selected in the networks. We modified the stability selection framework to use edge selection frequencies directly and combine them in the inferred network to ensure that only reproducible edges are included in the consensus. We demonstrated on synthetic data that our method generally led to slightly sparser networks while achieving much higher precision than any single method. We further applied the method to gut microbiome data from liver-cirrhotic patients and demonstrated that the resulting network exhibited a microbial guild that was meaningful in terms of human health.

ABSTRACT. Antibodies play a crucial role in the humoral immune response against health threats, such as viral infections. Training AI (Artificial Intelligence) models, for example to assist in developing sero-diagnostics or antibody-based therapies, requires building datasets according to strict criteria, to include as many standardized antibody sequences as possible. However, the available sequences are scattered across partially redundant databases and compiling them into a single non-redundant standardized dataset has hitherto remained a challenge.

Here, we present ABSD (AntiBody Sequence Database, https://absd.pasteur.cloud) which contains data from major publicly-available resources (abYbank, CATNAP-HIV, CoV-AbDab, GeneBank, IMGT, KABAT, OAS, PDB, PLAbDab, PairedNGS, SACS, SAbDab, UniProt...), creating the largest standardized, automatically updated and non-redundant (i.e., each antibody sequence stored in the database is unique) source of public antibody sequences for different species.

While ABSD contains over 1,350,000 antibody sequences today, trillions of them may circulate in the human population. This limitation is unlikely to be resolved anytime soon, but diversity might matter more than sheer number. In the article, we demonstrate that, at least regarding IGHV regions, our methodology does not seem to have introduced a strong bias in the selection of antibody sequences towards specific gene clusters, compared to a classic human repertoire.

When training deep learning models, the uniqueness and representativeness of the input data is likely essential for most applications. In this regard, ABSD will help mirror the human repertoire by providing, as broadly as possible and without bias, unique antibody sequences with realistic proportions.

Finally, ABSD is a dynamic and adaptive database, designed for automatic updates and easy upgrades. This user-friendly and open website enables users to generate lists of antibodies based on selected criteria and download the unique sequence pairs of their variable regions.

ABSTRACT. The Pfam protein families database is a comprehensive collection of protein domains and families used for genome annotation and protein structure and function analysis (https://www.ebi.ac.uk/interpro/). This update describes major developments in Pfam since 2020, including decommissioning the Pfam website and integration with InterPro, harmonization with the ECOD structural classification, and expanded curation of metagenomic, microprotein and repeat-containing families. We highlight how AlphaFold structure predictions are being leveraged to refine domain boundaries and identify new domains. New families discovered through large-scale sequence similarity analysis of AlphaFold models are described. We also detail the development of Pfam-N, which uses deep learning to expand family coverage, achieving an 8.8% increase in UniProtKB coverage compared to standard Pfam. We discuss plans for more frequent Pfam releases integrated with InterPro and the potential for artificial intelligence to further assist curation. Despite recent advances, many protein families remain to be classified, and Pfam continues working toward comprehensive coverage of the protein universe.

ABSTRACT. MethMotif (https://methmotif.org) is a publicly available database that provides a comprehensive repository of transcription factor (TF)-binding profiles enriched with DNA methylation patterns. Since its inception in 2019, the platform has evolved to incorporate expanded datasets and advanced functionalities, deepening our understanding of context-specific TF functions. In its 2024 release, MethMotif expands its initial collection from 509 to over 700 position weight matrices (PWMs), all annotated with DNA methylation profiles. A key advancement of this update is the segregation of TF-binding motifs based on cofactors and DNA methylation status, allowing researchers to explore how gene ontology (GO) annotations and TF target genes can differ under varying cofactor contexts. MethMotif now supports two additional species: Mus musculus and Arabidopsis thaliana, broadening its applicability for comparative and translational research. By incorporating cofactor-based binding motifs, methylation profiles, and precomputed GO enrichments, MethMotif stands out as the first and only TF-binding motif database to integrate context-specific PWMs with epigenetic information, thus enabling deeper insights into the regulatory mechanisms governing gene expression.

ABSTRACT. Antibiotic resistance (ABR) is a major global public health issue, recognized as an urgent priority by international institutions. Especially regarding the emergence and the global dissemination of multidrug-resistant bacteria (MDRB) and antibiotic resistance genes (ARGs), as they can spread widely across humans, animals and the environmental domains, transcending borders. We present ABRomics-analysis, an online platform designed for tracking MDRB and ARGs from bacterial genomes, with metagenomes analyses planned for the future. This platform aims to provide a tool for the surveillance and the research in antimicrobial resistance in a One Health context. ABRomics-analysis offers a user-friendly interface for uploading and managing sequencing data, launching standardized bioinformatics workflows, and exploring the ABR content of analyzed samples. Once samples have been uploaded, a default workflow is automatically launched, which includes (for whole-genome sequencing data) quality control of the raw data, species identification, contamination checking, sequence typing, genome assembly, genome annotation, plasmid typing, antibiotic resistance gene detection, and virulence gene detection. Users will also be able to perform custom analyses to answer more specific ABR-related questions, such as core-genome multilocus sequence typing (cgMLST) in a future platform version. Once the analysis is complete, an online report summarizes the sample metadata and the results from each step of the analysis. Users can download individual output files (e.g., assembled genome FASTA files, ABR gene detection tables) or retrieve multiple results for cross-sample analyses. For a comprehensive view of ABR gene transmission and to foster broad collaboration, the ABRomics-analysis database enables users to explore all available data through a variety of tools: a database overview with a list of filters, including sample type, spatial and temporal metadata, antibiotic classes associated with detected ABR genes, species, and sequence type detected; an interactive world map that can be filtered to show the localization of collected samples; and an anonymous contact system that facilitates communication and collaboration among users. ABRomics-analysis is accessible at https://analysis.abromics.fr/

ABSTRACT. WHOOPER is an intuitive web application designed to facilitate the analysis of protein co-occurrence across taxa. By integrating HHblits and HMMER, it enables efficient detection of remote homologs across extensive proteomic datasets. Its results interface combines an original data representation with advanced dynamic filtering functionalities to enable real-time exploration and exploitation of alignment data. Addressing key limitations of existing tools, WHOOPER streamlines workflows for studying protein prevalence, co-occurrence, and phylogenetic patterns, supporting seamless export for downstream evolutionary and functional genomics research.

ABSTRACT. We introduce InMoose, an open-source Python environment aimed at omic data analysis. We illustrate its capabilities for bulk transcriptomic data analysis. Due to its wide adoption, Python has grown as a de facto standard in fields increasingly important for bioinformatic pipelines, such as data science, machine learning, or artificial intelligence (AI). As a general-purpose language, Python is also recognized for its versatility and scalability. InMoose aims at bringing state-of-the-art tools, historically written in R, to the Python ecosystem: ComBat, ComBat-Seq, limma, edgeR, DESeq2. InMoose focuses on providing drop-in replacements for R tools, to ensure consistency and reproducibility between R-based and Python-based pipelines. The first development phase has focused on bulk transcriptomic data, with current capabilities encompassing data simulation, batch effect correction, and differential expression analysis and meta-analysis.

ABSTRACT. Spatial transcriptomics is a powerful method to study the spatial organization of cells, which is a critical feature in the development, function and evolution of multicellular life. However, sequencing-based spatial transcriptomics has not yet achieved cellular-level resolution, such that advanced deconvolution methods are needed to infer cell-type contributions at each location in the data. Recent progress has led to diverse tools for cell-type deconvolution that are helping to describe tissue architectures in health and disease. In this resource we describe the varied types of cell-type deconvolution methods for spatial transcriptomics, contrast their capabilities, and summarize them in a Shiny interactive table to enable more efficient method selection. Our tool gathers 67 cell-type deconvolution methods specifically developed for spatial transcriptomics. We categorize them with two classification systems depending on the use of single-cell resolved RNA-seq data and the mathematical framework used to formulate and solve the deconvolution problem. In addition, we gather 30 different items which detail across all the methods the expected input and output, data pre-processing, strategies to integrate scRNA-seq, image and coordinates, computational and biological validation with links to documentation. Our application is the first extensive database dedicated to cell-type deconvolution and provide a true mean to describe cell-type deconvolution landscape for spatial transcriptomics. Cell-type deconvolution is undergoing incredible advances, providing new insights into the organization of cellular communities from spatial transcriptomics data. The sheer number of methods has made difficult to choose a method and identify improvement venues. Thanks to the application’s interactive filtering and search bar, method developers and users can fetch for families of methods with given characteristics. On the methodological development side, our app gives a comprehensive overview of the mathematical models already used and considerations on how to benchmark such methods regarding simulation, datasets and metrics commonly used, leading to better targeting areas of improvement. On the user side, depending on tissue type, input quality, required output and programming fluency, our app allows to identify potential methods of interest tailored to one’s project. Furthermore, as cell-type deconvolution method development is an active field of research, we provide a form to maintain our tool up to date with recent development in the field. In conclusion, our app offers a comprehensive and interactive resource for selecting and understanding cell-type deconvolution methods for spatial transcriptomics. It enables efficient method selection tailored to specific research needs. As the field advances, our app supports both developers and users in improving and applying these methods, with ongoing updates to reflect the latest developments.