BIH SpaceHack project presented in Nature Genetics

The SpaceHack project driven by Dr. Naveed Ishaque, a BIH Group Leader in the Center of Digital Health, was recently covered in an editorial by Nature Genetics. In this article, we describe the importance of spatial transcriptomics and devise a strategy to reach an extensible set of guidelines for best practices regarding data analysis.

The importance of spatial transcriptomics

Spatial transcriptomics is a cutting-edge technique that plays a pivotal role in advancing our understanding of tissue architecture at the molecular level. It was announced as “Nature Method of Year” in 2020, as unlike traditional bulk or single-cell transcriptomic analyses, spatial transcriptomics allows researchers to investigate gene expression patterns within the spatial context of the tissue. This revolutionary approach not only enhances our comprehension of cellular heterogeneity but also provides critical insights into the intricate organization of cells within tissues. The significance of spatial transcriptomics lies in its ability to unravel the spatial distribution of gene expression, enabling researchers to decipher the unique molecular signatures of different cell types and their interactions within a tissue.

This spatial resolution is crucial for deciphering complex biological processes, such as development, immune responses, and disease progression, where the spatial organization of cells and their interactions are of paramount importance in normal development and disease. Capturing the spatial context of gene expression, researchers can gain a more comprehensive understanding of how cells communicate and coordinate their activities. This knowledge is particularly valuable for unravelling the mechanisms underlying diseases like cancer, where the spatial distribution of tumor cells and their interactions with the surrounding microenvironment are pivotal determinants of disease progression and treatment response.

Community driven benchmarking of computational tools

The success of spatial transcriptomics hinges heavily on the availability and effectiveness of computational tools for data analysis. Analysing vast amounts of spatial transcriptomics data demands robust computational methods that can accurately process, interpret, and visualize the intricate spatial patterns of gene expression. There has been an explosion in the number of computational tools to analyse spatial transcriptomics data. Evaluating and refining these computational tools, with a particular focus on community-driven community benchmarking, is crucial. A major limitation in the current paradigm of benchmarking is that they are often difficult to reproduce or extend with new tools and data. However, community-driven benchmarking ensures that the tools are not only reliable and reproducible but are also aligned with the evolving needs of the scientific community. This collaborative approach facilitates the establishment of benchmarks and standards, fostering a shared understanding of the strengths and limitations of different tools.

The SpaceHack project

The SpaceHack project aims to investigate the state of computational tools for spatial transcriptomics in a hackathon setting. The project runs as part of BioHackathon Germany, and has thus far attracted over 100 in-person and virtual participants in 2022 and 2023.

In 2023 the focus of SpaceHack was to implement computational tools to identify tissue domains and niches, apply them to reference datasets across technologies (e.g. Visum, Xenium, Slide-seq, MERFISH) and tissues (e.g. brain, heart, liver) and quantitatively evaluate their performance.

Example of spatial domains in the mouse brain dorsal lateral prefrontal cortex. The aim of domain identification is to identify the ground truth tissue domain (left) using computational methods. However, the performance of these method is highly variable. Image adapted from Dong et al. 2022 https://doi.org/10.1038/s41467-022-29439-6.

During SpaceHack we implemented over 25 datasets, 25 computational methods and 15 evaluation metrics by a team spread over three continents including participants from Seattle (Allen Institute for Brain Science) to Singapore (Genome Institute of Singapore). All the combinations of data x method x metrics were run in the OMNIBENCHMARK community benchmarking framework from the group of Prof. Mark Robinson at the UZH. We plan to share the final results of the SpaceHack project later this year.

The OMNIBNECHMARK framework. The OMNIBENCHMARK framework builds on establishing various modules as components of a benchmark: datasets, methods and method. Evaluation metrics can be interactively explored using a dashboard. Image adapted from https://omnibenchmark.org/about/.

An integral part of SpaceHack are the on-boarding lectures, where we regularly attract leaders in the field of spatial transcriptomics data analysis. In addition to presentation on data, methods and benchmarking experts, we hosted a talk from Dr. Teresa Zulueta-Coarasa from the EBI BioImageArchive describing the current state of affairs for data management and archival of spatial transcriptomics data.

The future of SpaceHack

The efforts of the SpaceHack participants have led to the generation of over 100,000 results. Our current focus is to summarize these results into analysis guidelines and recommendations for the life science community, and open up the benchmark to the community to allow others to easily test new tools, or run the implemented tools on their data.

We are committed to running SpaceHack again in 2024 in an inter-continental setting. In 2023 we struggled with central management across three continents and seven time zones. To address this issue, we plan to distribute the SpaceHack management across three in-person locations in the US (Harvard) and Australia (Adelaide) and Germany (Kassel) - please reach out to Naveed Ishaque if you are interested and would like to take part!