Access to ground-truth pseudobulk gene expression matrix

[tjchen020524]

Hi Decima team,

Thank you for making Decima and the accompanying documentation publicly available.

I’m working on a research project that extends sequence-based gene expression models by incorporating spatial microenvironment information, and I’m building on Decima while keeping pseudobulk gene expression as one of the model outputs.

I have a question regarding the ground-truth pseudobulk gene expression matrix used for training and evaluation.

I understand that multiple single-cell datasets were aggregated into pseudobulk samples and used as training targets for gene expression prediction. The API also exposes DecimaResult.ground_truth, which suggests that the true expression matrix can be accessed when available.

However, in the released W&B metadata and the supplementary data on Zenodo, I only see predicted expression values, and the AnnData .X matrix appears empty. I was unable to find a publicly available version of the ground-truth pseudobulk expression matrix in the documentation, tutorials, or the 'decima-applications' repository.

Could you clarify whether the ground-truth pseudobulk expression matrix is publicly available, and if not, what the recommended way is to obtain it?

Thanks very much for your time.

[MuhammedHasan]

The ground-truth pseudobulk data is not publicly available. See #15 for further details.

[avantikalal]

Hi @tjchen020524, we are looking into this and will get back to you in a few days.

[stephaniehicks]

Thank you @avantikalal! I appreciate you looking into this.