Help: Predicting Genetic Perturbation

[const-ae]

This is not actually an issue but rather a request for help :)

I would like to include Prophet in my benchmark of foundation models for predicting the effect of genetic perturbation on gene expression in cell lines. (https://www.biorxiv.org/content/10.1101/2024.09.16.613342).

Based on the preprint, I understand this is something Prophet can do; however, I don't understand which arguments to use when calling prophet.Prophet:

• For the cl_emb_path, I would use the provided cell_line_embedding and set ph_list = ["K562"].
• I don't understand what the ph_emb_path and target_phs should be for RNA-seq log-transformed counts.
• Similarly, iv_emb_path and target_ivs I am not sure how I can encode the genetic perturbations.

Any hints are appreciated. Alternatively, if you say that the software is in too early of a stage, that's also okay :)

Best,
Constantin

[AlejandroTL]

Hey,

Prophet is not initially thought to be applied to Perturb-seq experiments to predict transcriptome gene expression, which is what you want to do if I understand correctly. This is fully intentional. We are currently debating whether extending it in the future to do so. Also, if you check the datasets we use, there is no much single-cell and we don't even try to predict RNA-seq expression, we don't use the famous Replogle, Norman, etc.
If we decide to extend it to full transcriptomics we will let you know.

Thanks!

[const-ae]

Hi Alejandro,

thanks for the explanation. This is very helpful.

Feel free to close the issue if you want. I am looking forward to seeing where in which direction this project will develop :)

Best,
Constantin