Expanding scGPT Predictions to Additional Genes Beyond Tutorial Examples

[KristinTsui]

Thank you for the comprehensive tutorials on finetuning the scGPT model. I've been working through the Tutorial_Perturbation.ipynb and have successfully fine-tuned the model using the Adamson dataset. My objective now is to leverage this fine-tuned model to explore the perturbation effects on a broader set of genes of interest, beyond those directly covered in the tutorial examples.

In block 19 of the tutorial, under the "Evaluations" section, the plot_perturbation function seems to require that the query parameter match annotations found within adata.obs for conditions and adata.uns for mappings of top non-dropout differentially expressed genes.

Could you provide some guidance on how I could apply the model for prediction and visualization of perturbation effects on additional genes? More specifically, I would like to predict gene expression changes for a few gene present in adata.var.

Thank you in advance for any assistance or suggestions you can provide!