Results 

Congratulations to Sam Mestern from Western University Ontario for winning the MapMySpikes Data Challenge!  Sam had the most accurate algorithm and also submitted the tool that had the best rating and will be invited to present in an upcoming webinar… more details to come. 

Sam Mestern is a 3rd year Ph.D. student in the Inoue Lab at Western University. Sam is a patch-clamp electrophysiologist and computational researcher whose research aims to investigate state-dependent computation at the apex of the neuroendocrine stress axis. In his spare time, he works on hobby neuroscience projects and hangs out with his cats. Find him on Twitter at @smestern. 

 His submission, PatchOTDA (Patch Clamp Optimal Transport Domain Adaptation), is a small Python package and web interface that employs optimal transport-based domain-adaptation techniques to map the user sample onto the reference distribution. Once mapped, we can use a pre-trained classifier from the reference distribution to facilitate the labelling of the user sample. 

Also, a big thank you to the five other entrants to the challenge, including three who submitted algorithms, one who submitted a tool, and one who submitted both. Content from some of these teams will be included on Allen Brain Map later in 2024… more details to come.   

[Names are alphabetical]

• Xinyue Ma (McGill University): ElecFeX is a Matlab-based toolbox with a graphical user interface for extracting electrical properties from current-clamp recordings. It is open-source on GitHub with a detailed introduction in Cell Reports Methods. 

• Aayush Marishi (ETH Zurich): This approach used a hierarchical algorithm for classifying neuronal subtypes using patch clamp data, starting with supervised UMAP to cluster cells into broad categories, followed by recursive clustering and additional UMAP embeddings to refine subtypes. 
• David Zimmermann (Medical University of Innsbruck): Biophysical Essentials is a user-friendly, open-source one-click executable patch-clamp analysis software providing essential support for the entire workflow of transparent patch-clamp analysis, enhancing reproducibility in electrophysiology research. More details, including its extension for cell type classification, can be found at the project website.
  

Query data sets 

Three query data sets were included as part of MapMySpikes but kept anonymous during the challenge. These data came from the following manuscripts: 

• Query 1 – Hostetler et al 2023: “Genetically Defined Subtypes of Somatostatin-Containing Cortical Interneurons” includes a set of patch-clamp recordings on multiple genetic mouse lines each targeting known subsets of Somatostatin interneurons.   

• Query 2 – Scala et al 2021: “Phenotypic variation of transcriptomic cell types in mouse motor cortex” includes a second, smaller set of Patch-seq data as the CTKE reference data set collected at a different temperature, but mapped to transcriptomics cell types in motor cortex in the same way. 

• Query 3 – Wu et al 2023:  “Cortical somatostatin interneuron subtypes form cell-type-specific circuits” includes a set of patch-clamp recordings on a different set of genetic mouse lines targeting known subsets of Somatostatin interneurons. We thank Gord Fishell and his lab for providing electrophysiology features from this study for use in this challenge. 

 An updated MapMySpikes data file that includes correct cell to cell type mappings is now available at this link. For all three query sets, cell type assignments from the original manuscripts were used as a ground truth. Details of how cell types were matched between cell type taxonomies are described in the same file. 

About This Challenge

Despite the great utility of cell type mapping using gene expression, no user-friendly tool exists for data from other modalities, making the transfer of brain cell type knowledge from Allen Brain Map to scientists collecting non-omics data a labor-intensive process.  In particular, many studies include the use of patch clamp recording to characterize electrical properties of cells in different situations.  Having the ability to assign these cells to known cell types would aid in the interpretation of these data by, for example, providing a name to these cell types in common use from the literature and providing a set of commonly expressed ion channels or other genes in these cell types. 

Challenge: 1) create an algorithm that will accurately match patch-clamp recorded neurons from the three example data sets to the most likely cell types in mouse primary visual cortex (VISp) and/or primary motor cortex (M1), 2) present your mapping results as a user-friendly tool that links to the mapped cell types on the  Mouse Patch-seq VISp (VISp cells) and CTKE (M1 cells). 

Link to graphical representation of data challenge.

This challenge includes query data from three recent studies in mouse cortex which use patch clamp recording along with other methods to assign cells to transcriptomics-based cell types defined in recent work from the Allen Institute or BICCN (BRAIN Initiative Cell Census Network) references for the query studies will be released upon challenge completion).  Participants will be given the electrophysiology features and limited additional metadata for patch clamp-recorded cells from all three papers, but with information about cell type assignments and mouse transgenic lines omitted. Participants will also be given corresponding electrophysiology features from Patch-seq cells in matched studies of mouse VISp (Gouwens et al 2020) and M1 (Scala et al 2021), which include baseline data from the Mouse Patch-seq VISp and Cell Type Knowledge Explorer (CTKE) resources on brain-map.org, respectively.  

Entry Details

To be considered for the competition, please submit at this link: 1) files containing your mapping accuracy, 2) files containing your summarization of results, and 3) a file containing names and contact information of yourself and anyone on your team. Challenge datasets and additional information on file formats available in the MapMySpikes data file. 

If you have any difficulties with submission, please contact Rachel Hostetler.

Participants will be scored based on two components, with a winner selected for each category independently: 

1. Mapping accuracy: This winning algorithm for this category will be selected based on a quantitative assessment of accuracy to one of three transcriptomics-based cell type classifications using the provided input data alone. These classifications include (1) MET types in VISp (Gouwens et al 2020); (2) T-types in VISp (Gouwens et al 2020; originally from Tasic et al 2018); and (3) T-types in M1 (Scala et al 2021; originally from Yao et al 2021).  Cells aligned to either T-type will be considered accurately mapped if that T-type corresponds to the best matching MET type(s).  
2. Summarization of results: Creation of a user-friendly tool that integrates mapping results with content from similar cells in the Mouse Patch-seq VISp view and/or related cell type knowledge cards from the CTKE, and (optionally) additional information about cells from the query data set in a useful way. The definition of tool is intentionally left open, but some examples could include a website, an interactive table, or an R Studio or Jupyter notebook.  This winner will be selected by a panel of judges who will be grading on the following criteria: (1) inclusion of both query and reference data, (2) connections (via links or embedding) to relevant content on Allen Brain Map, (3) ease of use, (4) and functionality (e.g., searching, filtering, novel data upload, data visualization, etc.). 

Winners will have an opportunity to present their work at a to-be-scheduled webinar and will have their tool featured on Allen Brain Map (brain-map.org).  Other tools may also be included on Allen Brain Map and all participants agree to have their submissions publicly linked at Allen Brain Map.  

Useful links for participants:

• Mouse M1 data (underlying the CTKE)
• • Cell Type Knowledge Explorer: https://portal.brain-map.org/atlases-and-data/bkp/cell-type-knowledge-explorer  
  • Public codebase for CTKE: https://github.com/AllenInstitute/CTKE_viz/tree/main 
  • Electrophysiology data on DANDI for M1: https://dandiarchive.org/dandiset/000008 
  • Example location on AWS for a CTKE image: https://cell-types-images-production.s3.us-west-2.amazonaws.com/CCN202002013/VLMC_10X_nuclei_v3_Broad_UMAP2d.png 

• Mouse VISp data (underlying the Mouse Patch-seq VISp viewer) 

• • Mouse VISP patch-seq specimen viewer: https://knowledge.brain-map.org/data/1HEYEW7GMUKWIQW37BO/specimens 
  • Electrophysiology data on DANDI for VISp: https://dandiarchive.org/dandiset/000020 
  • Morphoelectric data (before patch-seq) on data from mouse VISp and human neocortex: https://celltypes.brain-map.org/data 
• MapMyCells: https://portal.brain-map.org/atlases-and-data/bkp/mapmycells  

Background 

The Cell Type Knowledge Explorer (CTKE) is a publicly accessible web application to browse data associated with the multimodal cell census and atlas of primary motor cortex, developed in close collaboration with the BRAIN Initiative Cell Census Network (BICCN) and described in detail in the BICCN flagship publication. The CTKE includes access to single-cell transcriptomic and epigenomic profiling of human, marmoset, and mouse primary motor cortex, with additional data sets assessing the spatial distribution, cell morphology (shape), and electrical properties of these cell types. All data powering the CTKE is provided to encourage cell exploration by scientists, students, and educators through the browser and linked resources. 

Allen Brain Map and BICCN have tools for allowing scientists to compare their gene expression data with cell types in motor cortex (Azimuth; accessible via the CTKE) and spanning the whole brain (MapMyCells).  These tools directly transfer (or “map”) cell type assignments to user-provided cells, also allowing a transfer of associated knowledge about that cell type.  For example, cells in motor cortex matching to Pvalb interneurons based on gene expression are also likely to be fast spiking and to have either basket or chandelier morphologies.  Similarly, putative spatial locations can be assigned to a cell collected from a dissociation protocol by mapping it to the whole mouse brain and then seeing where that cell is located using the Allen Brain Cell (ABC) Atlas.  Finally, by mapping cell type assignments to cells collected from individuals with Alzheimer’s disease (e.g., in SEA-AD) and other disorders, scientists can understand which cell types change and how they change in different states.  While other algorithms exist for mapping transcriptomics (and other ‘-omics’) data between reference cell types and query data, Azimuth and MapMyCells are web-based, making them accessible to researchers and students with a wide range of computational and biological competencies. 

This data challenge is supported by the National Institute Of Neurological Disorders And Stroke of the National Institutes of Health under Award Number U24NS133077