Single-cell RNA sequencing has proved an invaluable tool in biomedical research. The ability to survey the transcriptome of individual cells offers many opportunities and has already paved the way to many discoveries in both basic and clinical research. With the increasing amount of single-cell transcriptomic data available, including whole-organism single-cell transcriptomic atlases, biological databases face a challenge to integrate these data and make them easily accessible to their users. In this Short Paper, we describe how the fruit fly-specific database FlyBase is making use of single-cell RNA sequencing datasets to let fly researchers quickly know in which cell types genes are known to be expressed.
First introduced in 2009, single-cell RNA sequencing (scRNAseq) has become a powerful tool to investigate cell states and functions [1]. Since 2017, many studies have leveraged the technique in the fruit fly Drosophila melanogaster, leading to various new insights [2], and the number of scRNAseq datasets from flies is only expected to grow quickly in the coming years. Recently, a consortium of 40 fly laboratories, the Fly Cell Atlas project, completed and reported the first single-cell transcriptomic atlas of the entire adult fruit fly, sequencing more than half a million cells of 250 different types across 17 tissues [3].
FlyBase is the Model Organism Database (MOD) for all data related to Drosophila melanogaster [4]. It provides access to a wide range of scientific information either manually curated from the published literature or from high-throughput research projects. Expression data are a particularly important subset, and we strive to give access to as many resources as possible to allow fly researchers to find out in which tissues and at which developmental stages their gene of interest is known to be expressed. To that end, we already make use of several high-throughput projects such as modENCODE [5], FlyAtlas [6], and FlyAtlas 2 [7]. We now want to exploit available scRNAseq
the available Drosophila scRNAseq datasets; (b) get some information (metadata) about these datasets; (c) get a quick overview of the expression data from those datasets.
In this paper, we will focus on (c), and explain how we leverage the expression data provided by scRNAseq experiments to obtain and present a percell type summary of the expression data, so that users can gen an “immediate” answer (straight from the “Gene Report” page) to the following questions: What are the cell types in which a given gene is expressed? What is the proportion of cells of a given type in which the gene is expressed? What is the average expression level of the gene across all cells of that type?
An important design decision is that no scRNAseq dataset will ever be stored within FlyBase itself. Instead, we will only store some metadata as well as a simplified version of the gene expression data – only what is needed to cover the questions listed above. Users wanting to inspect the full data will be directed to a pre-existing external data store where the data will already be available.
obtained through the Single Cell Expression Atlas (SCEA), the EMBL-EBI resource for gene expression data at the single cell level [8]. We thus have a single data provider that is independent of all the individual research projects. The SCEA data curators process the original raw data, as deposited by the original authors in common data stores such as the Gene Expression Omnibus or Array Express (Figure 1, steps 1–2), according to a standard pipeline with common parameters that allows for cross-dataset comparisons. As an added benefit, the SCEA provides the processed data in a common format, independently of the actual scRNAseq method originally used (e.g., Smart-seq2 or 10⨉ sequencing).
cell type name, not knowing that this cell type is formally known under another name in the DAO (an example is “astrocyte”, for which the actual term in the DAO is “astrocyte-like glial cell”); 2. an annotation may refer to a cell state rather than, or in addition to, a cell type (an example is “plasmatocyte-prolif”, which was used in a dataset to annotate a cluster of proliferating plasmatocytes; the appropriate ontology term in that case is “plasmatocyte”, leaving the cell state aside); 3. an annotation may incorrectly refer to an organ or a tissue rather than to the cell types that make up this organ or tissue (an example was a cluster annotated as “dorsal vessel”, where the correct cell type name should have “cardial cell”); 4. an annotation may refer to several cell types (an example was “OPN innervating DA1, VA1 or DC3 glomerulus”; the correct term in such case is the closest cell type that encompasses all cell types in the cluster, in this instance “olfactory projection neuron”).
Lastly, there is one case where the original annotation cannot be converted to an ontology term: when the cell type is either not clearly identified (e.g. “btl-GAL4 positive cell, likely to be ovary cell”), or identified only for what it is not (e.g. “non-hemocyte”). Such clusters will be excluded from the expression data summarisation described below.
Of note, this “conversion” of the original annotations is not destructive: the original annotations from the upstream authors are at all time preserved. FlyBase curators merely add a new set of ontology-compliant annotations alongside the original annotations. Both sets of annotations are ultimately available on the Single Cell Expression Atlas.
In the process of converting the original annotations to DAO-compliant annotations, the need to update the DAO itself may arise, for mostly two reasons: a) a dataset reports a new cell type, which must be added to the ontology (an example was “adipohemocyte”, a subtype of plasmatocytes found in a scRNAseq analysis of the lymph gland [10]); b) the ontology is found to lack some “grouping terms”, useful to offer a better qualification of some clusters (an example was the addition of “endocrine cell of the ring gland”, to encompass the various types of endocrine cells found in the ring gland). FlyBase data curators then work alongside FlyBase ontology editors to update the DAO as needed and the newly added terms are used to annotate the scRNAseq clusters.
In some cases, this process of updating the DAO to better annotate a dataset may actually start before the dataset is even published, through a direct collaboration between FlyBase and the dataset’s authors. This was notably the case of the Fly Cell Atlas dataset [3].
processed by the SCEA standard pipeline, the most important output is the Normalised Expression Matrix: for any single dataset, this is a large matrix of n cells by m genes, with n the total number of cells in the dataset and m the total number of identified genes; each matrix cell contains the normalised reads count for a particular gene in a particular cell. This matrix is accompanied by the Experiment Design Table, which contains annotations for each single cell (notably the cell type annotations, both as submitted by the original authors and as corrected by the FlyBase curators as described above).
From those two files, we proceed to what we call the per-cell type summarisation of the gene expression data (Figure 1, step 7). For each gene and for each identified cell type, we extract three measures: a) the cluster size, that is the number of cells annotated with that cell type; b) the spread or extent of expression, the proportion of cells in the cluster in which the gene is expressed (the number of cells in the cluster with a non-zero reads count for that gene, divided by the cluster size); and c) the average expression of the gene in positive cells (the total number of reads across all cells in the cluster, divided by the number of cells with a non-zero count). Those measures are then stored in the Chado database at the heart of FlyBase, from where they can be exploited on the FlyBase website.
The summarised data are also slated to be used by Virtual Fly Brain [11]. To that effect, a subset of the data, corresponding to the cell types present in the adult brain only and to genes with an extent of expression of more than 0.9 (genes that are expressed in more than 90% of cells of a given type), is loaded from the Chado database to the Neo4j graph database (Figure 1, step 8) that powers the Virtual Fly Brain website.
exploited on the FlyBase website in several incremental steps. The first step was realised for the 2022_03 release of FlyBase in June 2022. In this release, a cell type ribbon was added to the Gene Report page (Figure 2A). The ribbon lists 22 high-level cell types, each of those being coloured according to the extent of expression of the current gene in cells of that type. Because several more precise cell types are agglomerated into a single high-level cell type, a list of all the precise subtypes, with their corresponding extent of expression, is given in a tooltip box shown when the mouse pointer is left hovering over a ribbon cell (Figure 2B). To avoid cluttering that tooltip, cell types where the extent of expression is less than 1% are filtered out.
At present, the cell type ribbon is only fed with the data from the Fly Cell Atlas project [3]. In a second step and as more datasets will be accumulated into FlyBase, we will update the ribbon to display consolidated data from all the available scRNAseq datasets.
Lastly, in a future release we will add a new type of graph to the “high-throughput expression
[1] Haque, A., Engel, J., Teichmann, S.A.,
Lönnberg, T., A practical guide to singlecell RNA-sequencing for biomedical research and clinical applications. Genome
Med 2017, 9, 1–12. [2] Li, H., Single-cell RNA sequencing in
Drosophila : Technologies and applications.
WIREs Dev Biol 2021, 10. [3] Li, H., Janssens, J., De Waegeneer, M.,
Kolluru, S.S., et al., Fly Cell Atlas: A single-nucleus transcriptomic atlas of the adult fruit fly. Science 2022, 375, Figure 2: Graphical display of summarised [4] eLaabrkki2n4,3A2.., Marygold, S.J., Antonazzo, G., expression data on FlyBase. (A) The cell type Attrill, H., et al., FlyBase: updates to the ribbon. Each cell in the ribbon corresponds to Drosophila melanogaster knowledge base. one of the main Drosophila cell types. Each cell is Nucleic Acids Res 2020, 49, D899–D907. coloured depending on the fraction of cells of [5] Graveley, B.R., Brooks, A.N., Carlson, that type in which the current gene is expressed J.W., Duff, M.O., et al., The developmental according to the Fly Cell Atlas dataset. (B) When transcriptome of Drosophila melanogaster. the mouse pointer is hovering over one cell, a Nature 2011, 471, 473–479. tooltip shows the list of cell subtypes in which [6] Chintapalli, V.R., Wang, J., Dow, J.A.T., the current gene is expressed, along with the Using FlyAtlas to identify better Drosophila proportion of cells expressing the gene in each melanogaster models of human disease. Nat subtype. (C) Mockup of a more complete Genet 2007, 39, 715–720. graphical representation of summarised [7] Leader, D.P., Krause, S.A., Pandit, A., expression data that is planned for a future Davies, S.A., Dow, J.A.T., FlyAtlas 2: a release of FlyBase. For each cell type, this graph new version of the Drosophila melanogaster will display both the proportion of cells of that expression atlas with RNA-Seq, miRNASeq and sex-specific data. Nucleic Acids type in which the gene is expressed (left) and the Research 2018, 46, D809–D815. average expression level in all cells that do [8] Papatheodorou, I., Moreno, P., Manning, J., express the gene (right). Fuentes, A.M.-P., et al., Expression Atlas data” section of the Gene Report page. The exact update: from tissues to single cells. Nucleic modalities of this new graph are yet to be Acids Research 2019, gkz947. determined, but il will display both the extent of [9] Costa, M., Reeve, S., Grumbling, G., expression and the average expression level for a Osumi-Sutherland, D., The Drosophila given gene across high-level cell types (Figure anatomy ontology. Journal of Biomedical 2C). Semantics 2013, 4, 32. [10] Cho, B., Yoon, S.-H., Lee, D., Koranteng,
F., et al., Single-cell transcriptome maps of 5. Acknowledgements myeloid blood cell lineages in Drosophila. Nat Commun 2020, 11, 4483.
This work was supported by joint grants [11] Milyaev, N., Osumi-Sutherland, D., Reeve, BB/T014008/1 (FlyBase) and BB/T014563/1 S., Burton, N., et al., The Virtual Fly Brain (EMBL-EBI) from the UK Biotechnology and browser and query interface. Bioinformatics 2012, 28, 411–415.