Exploring bin2cell for Visium HD analysis

I discuss the bin2cell manuscript (https://doi.org/10.1093/bioinformatics/btae546) which covers Python-based software for the genomic analysis of Visium HD data.

Transcript

1. Bin2cell reconstructs cells from high resolution Visium HD data Polański et al. DOI: 10.1093/bioinformatics/btae5460.1101/2023.11.26.568752 Presented by: Nick Eagles 2024/10/09
2. Manuscript Overview - bin2cell operates on 2um bin resolution (highest offered by Visium HD) - Segments nuclei on H&E image then pools together bins that overlap cellular segmentations - Result is AnnData with genes as features and cells as observations
3. Workflow Overview
4. Assigning bins to individual cells
5. bin2cell accurately places cell types
6. bin2cell: advantages over 8um-bin analysis - bin2cell-constructed cells can be more confidently annotated - bin2cell-constructed cells have more expressed genes
7. Accuracy of cell type and placement - Bin2cell improves accuracy of cell-type calls - 8μm bins predict cell types outside of tissue 8μm bin2cell
8. bin2cell adjusts for novel technical effect Adjusts for novel technical effect: biased counts by row or column
9. Expansion strategy: nuclei to cells - First, nuclei are segmented on H&E images - Two possible expansion strategies are used to estimate cell bounds - Expanding by a fixed distance (e.g. 2 bins or 4um) - Expanding to satisfy cell-nucleus volume ratio (e.g. 4) - Lower expansion gave better annotation confidence but biased towards higher-gene-expression cell types - Optimal expansion strategy depends on density of cells and expression levels of particular cell types