Biology subjects

Puig-Barbe, A.

Publications and source records attributed to Puig-Barbe, A..

6 recordsLinked to original sources

bioRxiv2026

AI semantics for biomedical data integration

Researchers increasingly need to explore hypotheses that span multimodal data across different scales, organisms, and domains. In practice, this requires connecting knowledge across fragmented databases with incompatible APIs and heterogeneous annotation practices. Large language model (LLM) agents can automate this data integration process, but grounding LLM agent outputs in scientifically correct sources of truth remains a significant challenge. Here we describe our deployment of a novel AI semantics workflow using LLM agents to enable scalable data integration, grounded in biological knowledge in the form of ontologies. Our workflow comprises (1) a multi-agent system curating scientific knowledge across ontologies using the Ontology Lookup Service (OLS) as grounding; (2) an LLM embedding service to enable interoperability between scientific databases by mapping ontology terms; and (3) GrEBI, a knowledge graph and Model Context Protocol (MCP) server enabling LLM agents to conduct cross-cutting, multi-omic biomedical queries.

bioinformatics↗

bioRxiv2026

Organism-scale annotation with Pan-human Azimuth

Single-cell atlases now span many human tissues, but inconsistent annotations across studies limit their utility as a unified reference. We introduce Pan-human Azimuth, a supervised neural network that maps human cells from diverse tissues and datasets onto a single hierarchical organism-scale typology. Developed through NIH HuBMAP, the model is trained on a uniquely curated corpus designed to maximize diversity across tissues and technologies while enforcing uniform, interpretable annotations and stringent quality control. The use of a single organism-wide reference enables us to map tens of millions of cells in the Tabula Sapiens and scBaseCamp repositories, perform cross-tissue comparisons across thousands of samples, and identify striking tissue specialization among fibroblast states. Pan-human Azimuth naturally extends to annotating spatial transcriptomic data, recovering canonical kidney cortical structures and distinguishing glomerular states consistent with expert pathology. We release Pan-human Azimuth alongside cloud, R, and Python interfaces to facilitate standardized organism-wide single-cell analysis.

genomics↗

bioRxiv2025

Exploring endothelial cell environments across organs in spatially resolved omics data

Endothelial cells are ubiquitously present in the human body and line the luminal surface of blood and lymphatic vessels. The oxygen-dependence of cells impacts their proximity to blood vessels, and consequently, to endothelial cells depending on their functional properties and priorities. This paper presents cell-to-nearest-endothelial-cell distance distributions for various cell types using 399 spatially resolved omics datasets from 14 studies comprising 12 tissue types with a total of 47,349,496 cells. Additionally, we developed an open-source web-based interactive tool, Cell Distance Explorer, that allows researchers to interactively visualize cell graphs and linkages in 2D and 3D datasets. Finally, we present a hierarchical neighborhood analysis focused on the endothelial cell neighborhoods in small and large intestine datasets. This paper provides an open-access resource (datasets, tools, and analyses) to characterize and compare cell distances and cell neighborhoods in spatially resolved omics data.

bioinformatics↗

bioRxiv2025

Constructing and Using Cell Type Populations of the Human Reference Atlas

The human body contains [~]27-36 trillion cells of up to 10,000 cell types (CTs) within a volume of [~]62-120 liters (males) and 52-89 liters (females). The Human Reference Atlas (HRA) v2.3 provides a quantitative 3D framework of CTs across 73 reference organs and 1,283 3D anatomical structures (ASs). The HRA Cell Type Population (HRApop) effort has quantified CTs per AS using high-quality single-cell datasets processed through scalable, reproducible workflows and cell type annotation (CTann) tools. HRApop v1.0 includes reference CT populations for 73 ASs (112 when sex-specific) using 662 datasets spatially registered to 230 locations across 17 organs (31 when sex-specific). For 558 single-cell (sc-)transcriptomics datasets (11,042,750 cells), CTs and biomarker expressions were computed using Azimuth, CellTypist, and popV. To test generalizability, 104 sc-proteomics datasets (16,576,863 cells) were integrated. In total, HRApop includes 27,619,613 cells and serves as a healthy reference for researchers aiming to elucidate mechanisms underlying cellular interactions, FTU operations, and cellular and tissue level disease progression, which may facilitate advancements in basic discovery and lead to new therapeutic strategies.

bioinformatics↗

bioRxiv2024

A general strategy for generating expert-guided, simplified views of ontologies

Annotation of biomedical entities with widely used, well-structured ontologies and ontology-aware tools ensures data and analyses are Findable, Accessible, Interoperable, and Reusable (FAIR). Standardized terms with synonyms support lexical search, while ontology structure enables biologically meaningful grouping of annotations, such as by location and type. However, ontologies serving diverse communities are often more complex than needed for specific applications, creating barriers to adoption by researchers and resource developers. For example, cell atlases often attempt simplifications by manually building term hierarchies linking to cell type and anatomy ontologies, but these may include relationship types unsuitable for grouping annotations. We present tools for validating human expert curated term hierarchies, developed in two human reference atlas projects, against ontology structures. The tools provide tabular statistics plus graphical views of matching and non-matching terms and relationships to support discussion and conflict resolution. The HuBMAP Human Reference Atlas (HRA) effort is used to validate the approach and tools, and the Human Developmental Cell Atlas is featured as a use case.

bioinformatics↗

bioRxiv2019

Homo- and heterodimerization of bHLH transcription factors balance stemness and bipotential differentiation

Multipotent adult stem cells balance self-renewal with differentiation into various cell types. How this balance is regulated at the transcriptional level is poorly understood. Here we show that a network of basic Helix-Loop-Helix (bHLH) transcription factors controls both stemness and bi-potential differentiation in the Drosophila adult intestine. We find that homodimers of Daughterless (Da), homolog of mammalian E proteins, maintain self-renewal of intestinal stem cells (ISCs), antagonising the Enteroendocrine fate promoted by heterodimers of Da and Scute (Sc, homolog of ASCL). The HLH factor Extramacrochaetae (Emc, homologous to Id proteins) promotes absorptive differentiation by titrating Da and Sc. Emc prevents the committed absorptive progenitor from de-differentiating, underscoring the plasticity of these cells. Switching physical interaction partners in this way enables the active maintenance of stemness while priming stem cells for differentiation along two alternative fates. Such regulatory logic is likely operative in other bipotent stem cell systems.

developmental biology↗