- Check out our new preprint on STARIT for creating image-based tensor representations of molecular resolution spatial transcriptomics data that can be combined with deep learning computer vision models to recover cell-states based on subcellular heterogeneity: www.biorxiv.org/content/10.6... 🧵👇
- Conventional analysis methods to identify cell-types and states in molecular resolution spatial transcriptomics data rely on tallying the number of molecules per gene per segmented cell, overlooking subcellular heterogeneity that can play important functional roles in defining cell states.Dec 22, 2025 19:10
- We developed STARIT (Spatial Transcriptomics As Rasterized Image Tensors) as an alternative way to represent molecular resolution spatial transcriptomics data to incorporate subcellular molecular organization.
- Using simulated data, we demonstrate that STARIT can distinguish transcriptionally distinct cell-types and further separates cell states based on subcellular transcript localization, which conventional gene count analysis fails to capture.
- Using real osmFISH spatial transcriptomics data, we demonstrate how STARIT can identify comparable cell-types to conventional gene count clustering analysis, with analogous differential feature analysis to give comparable insights to differential expression analysis.
- Using real bacterial-MERFISH spatial transcriptomics data, we demonstrate how STARIT can delineate rotational variation not distinguishable by conventional gene count analysis.
- Overall, we hope that STARIT will enhance the identification of novel cell-states by providing a structured framework to represent molecular resolution spatial transcriptomics data in a way that captures subcellular heterogeneity. Available as a Python package on GitHub: github.com/JEFworks-Lab...
- Congrats to @deevelazquez.bsky.social for leading this work 👏 🥳 🎉 Dee is applying to PhD programs so if you’re recruiting graduate students, definitely check out STARIT. Preprint: www.biorxiv.org/content/10.6... Software: github.com/JEFworks-Lab...
