TL;DR
Researchers have created scRetinaDB, a comprehensive, publicly accessible database that consolidates single-cell and spatial omics data from retinas across 17 species. This resource aims to overcome the fragmentation of existing data, enabling deeper cross-species comparative analyses of retinal development, disease, and regeneration.
What Problem Does This Paper Address?
The retina, crucial for vision, is implicated in various diseases. Recent advances in single-cell RNA sequencing (scRNA-seq), single-cell ATAC sequencing (scATAC-seq), and spatial RNA sequencing (spRNA-seq) have generated a wealth of data on retinal tissues. However, this valuable data is scattered across different public repositories, often with inconsistent processing. This fragmentation hinders comprehensive, integrative analyses, especially for cross-species comparisons, which are vital for understanding conserved mechanisms and identifying disease-specific insights. scRetinaDB directly addresses this by providing a unified, standardized platform for accessing and analyzing this diverse retinal omics data.
Methods
The authors developed scRetinaDB by systematically collecting and processing data from public repositories like GEO, GSA, and ArrayExpress, supplemented by literature searches. They employed standardized pipelines for data processing, quality control, normalization, and dimensionality reduction for each data modality (scRNA-seq, scATAC-seq, spRNA-seq). Key analytical approaches include:
- Integrative scRNA-seq analysis: For each species, scRNA-seq datasets were integrated using the Harmony algorithm to correct for batch effects, preserving biological variation. This allowed for the construction of species-specific retinal cell atlases.
- Cell type annotation: A novel three-step method was developed, utilizing literature-curated markers and calculating "marker power" and "joint power" based on AUC values to assign cell identities to clusters.
- Cell-cell communication analysis: The LIANA package was used to infer interactions between cell types, leveraging established ligand-receptor databases.
- Gene regulatory network inference: For human and mouse data, pySCENIC was employed to infer gene regulatory networks and assess transcription factor activity.
- Database implementation: A three-tier web architecture (frontend, backend, database) was used, with Vue.js for the interface, Django for backend logic, and MySQL for data storage.
Datasets
scRetinaDB integrates a massive collection of retinal omics data:
- scRNA-seq: Over 2.79 million cells from 453 datasets across 34 studies, spanning 17 species and 27 biological conditions (development, disease, regeneration).
- scATAC-seq: 107 datasets from human and mouse retinas.
- spRNA-seq: 18 datasets from human and mouse retinas.
The database covers a wide range of species, including human, macaque, mouse, zebrafish, chick, cow, pig, sheep, ferret, opossum, squirrel, peromyscus, rhabdomys, marmoset, tree shrew, lizard, and shark.
Key Insights and Ideas
- Data fragmentation is a major bottleneck: The paper strongly emphasizes that scattered, inconsistently processed data severely limits the potential of cross-species retinal research. scRetinaDB directly tackles this by providing a unified resource.
- Cross-species conservation and divergence are key: By integrating data from multiple species, scRetinaDB enables the identification of both conserved retinal cell types and functions, as well as species-specific adaptations and disease mechanisms.
- Multi-modal data integration is powerful: Combining scRNA-seq, scATAC-seq, and spRNA-seq data within a single platform allows for a more holistic understanding of retinal biology, from gene expression to chromatin accessibility and spatial organization.
- A user-friendly interface is crucial for adoption: The database features four main modules (Browse, Search, Analysis, Download) designed for intuitive exploration and utilization by researchers.
Key Results
- scRetinaDB hosts over 3.6 million retinal cells from 578 datasets across 17 species, 8 platforms, and 43 biological conditions.
- Species-specific retinal cell atlases were constructed by integrating scRNA-seq data, revealing detailed cell type compositions and subtypes. For example, the human retinal atlas identified 16 distinct cell populations.
- The database provides tools for cell type annotation, cell similarity analysis, cross-species gene expression queries, and spatial expression pattern exploration.
- The integrated human retinal atlas revealed distinct cellular compositions in normal, embryonic, age-related macular degeneration, and autoimmune retinopathy conditions, consistent with known pathology.
Limitations and Future Work
- Data coverage: While extensive, the database's coverage of all species and data modalities is not uniform. Human and mouse retinas are the most represented.
- Continuous updates: The authors are committed to continuously updating scRetinaDB with new data and modalities.
- Future directions: Future work includes incorporating additional retinal data modalities, such as single-cell DNA methylome sequencing, and developing further analysis tools. The datasets are also envisioned as valuable training resources for deep learning applications.
Summary updated