Unifying Omics Hackathon Series
Bridging scverse and Cell Painting communities at the intersection of image-based profiling, phenomics, and single-cell perturbation analysis.
About the event
Recent advances in Cell Painting and image-based profiling have created an emerging overlap of biological questions addressed by separate computational ecosystems. This hackathon brings together the scverse, CytoData Society, and EU-OPENSCREEN communities for three days of hands-on development at the MDC Berlin, built around three concrete tracks spanning Cell Painting analysis with scverse tooling, single-cell perturbation tools, and scalable model training.
We explicitly welcome analysts and biological scientists experienced in Cell Painting workflows alongside software developers, computational biologists, and ML researchers. The focus is high-throughput image-based profiling of cultured cells (Cell Painting, JUMP-CP, EU-OPENSCREEN), not clinical histopathology, H&E/IHC, or tissue-level imaging. Domain expertise is as critical as technical development: the goal is reusable open-source tools at the intersection of Cell Painting, single-cell perturbation analysis, and phenomics.
For Cell Painting practitioners: the scverse ecosystem provides standardised data structures, composable analysis workflows, and community best practices that have transformed single-cell transcriptomics. The single-cell best practices book illustrates the kind of streamlined, interoperable analyses we could build for image-based profiles, enabling seamless integration with perturbation data. In particular, we believe that the recent engineering efforts made by the tools towards being able to routinely handle large datasets (10+ millions of observations) can be leveraged for Cell Painting.
For ML researchers: Cell Painting generates phenotypic fingerprints of millions of cells across thousands of compounds, making it one of the richest high-throughput biological datasets available. Two good entry points into the field are the reviews by Caicedo et al. and Serrano et al..
Hackathon tracks
Closer to the event, we will publish a set of actionable tasks grouped into these six workstreams. You will later decide which task fits your expertise and interests best. Participants will group around these tasks and then work as a team toward a concrete, shippable outcome over the three days. Workstreams are topically distinct but not siloed: cross-workstream collaboration is encouraged wherever tools or data overlap.
Connecting typical CellProfiler workflows to the scverse ecosystem: running CellProfiler ourselves to work out which export settings carry real downstream value, specifying a single-cell AnnData representation and building the reader for it, adding a CellProfiler reader to spatialdata-io for images and masks, and writing an export plugin that skips the conversion step altogether.
@timtreisBringing optical pooled screening into scverse, where a phenotypic imaging round is paired with several rounds of in-situ sequencing. Deciding how to represent sequencing cycles and their registration, what breaks at plate scale, and what the ecosystem offers once images, masks and the per-cell table live in one object.
@LucaMarconatoAttaching defensible confidence estimates to predictions made from morphological profiles. Comparing Bayesian and conformal approaches on profiles stored in AnnData, testing whether their uncertainties survive compound- and site-level splits, and asking how an uncertainty should be exposed to downstream users.
@hanyangiiAligning Cell Painting profiles from LINCS and EU-OPENSCREEN with matched single-cell transcriptomics, using drug response as a supervision signal. Asking whether the two modalities are complementary or redundant, and which gene programs are encoded in a morphological phenotype.
@ArkkienkeliExtending the scverse single-cell image toolkit scPortrait towards generative modelling of cell morphology, interactive visualisation of image-based phenotypes, and agentic image-processing workflows.
@nik-asTesting how far the single-cell perturbation stack carries over to morphological profiles. Triaging which pertpy capabilities transfer, benchmarking the competing measures of perturbation strength against each other at both single-cell and well level, and pinning down which transcriptomic assumptions quietly break on image-derived features.
@timtreisProgramme
Wed Sep 2
Thu Sep 3
Fri Sep 4
The team
Collaborating organisations
Made possible by
Funding, venue & organisation
Participate
Whether you are an analyst or biological scientist experienced in Cell Painting, a software developer, computational biologist, or ML researcher: if you work at the intersection of Cell Painting and single-cell biology, this event is for you. Apply below or reach out with any questions.
Zulip
LinkedIn