An approach to generate a cell-type-enriched ChromBPNet model

[Y-ogativity]

Dear ChromBPNet developers,

Thank you for developing such an advanced deep-learning model that is user-friendly.

I would appreciate your feedback on the feasibility of my approach to generating a cell-type-enriched ChromBPNet model.

Specifically, I am interested in a ChromBPNet model that predicts counts and profiles in photoreceptor-enriched open chromatin regions (OCRs) to extract the unique sequence grammar of photoreceptor cells. The photoreceptor-enriched OCRs were generated by subtracting OCRs from the other OCRs of the other cell types (e.g., bipolar cells, etc.) in single-cell ATAC-seq data of retinas. The photoreceptor model will be trained only with photoreceptor-enriched OCRs. The bias model is trained with all photoreceptor OCRs, and nonpeak regions are defined with photoreceptor OCRs BEFORE the subtraction.

I would appreciate any comments and feedback on my approach.

Thanks,
Yohei Ogawa