Random thought, why no one is making seq based models for epigenetic code from single cell datasets. I get that we lack modalities, I.e. scCUT&TAG is super hard, but what about EPI-CyTOF? 30+ histone mods/TFs each experiment #DL #genomics
Comments
Log in with your Bluesky account to leave a comment
Yeah you are absolutely right. I'll rephrase it as it wasn't clear, my bad. What i meant is all baseline models bassnet/enformer/borzoi and so on are trained on different cell types with epigenetic data on bulk measures (ChIPseq, DNAse, ...). I was wondering whether re-training them w/ data
from single cell type at single cell level would be advantageous for modelling variability across cells. Though, the sparsity might force us to aggregate them to pseudobulk and we are back to square one. Alternatively one could approach the problem like scooby and instead pair scRNAseq
w/ EPI-CyTOF to predict single modalities to single cell levels. Or again, approach it like decima and first tune on scEpigenomics data and then tune again with scRNAseq like they did to predict expression. Hope i've made myself clear, i might be completely wrong or its just not worth it.
Comments
https://www.biorxiv.org/content/10.1101/2023.07.26.550634v1
https://www.biorxiv.org/content/10.1101/2024.10.09.617507v1
https://www.biorxiv.org/content/10.1101/2024.09.19.613754v2