Questions and Feedback on GSFA Tutorial and Functions

[hanhuajun40]

Hi, Xin,

I attempted to replicate the results from your GSFA tutorial using LUHMES CROP-seq data but encountered a few issues.

1. I couldn't reproduce the results shown in the attached dotplot_beta_pip.pdf file for dotplot_beta_PIP.
   dotplot_beta_pip.pdf
2. I noticed that several functions mentioned in the tutorial, such as complexplot_perturbation_factor, make_gibbs_res_tb, plot_pairwise.corr_heatmap, and complexplot_gene_factor, are not available in the GSFA package. Could you provide the code for these functions or update the package to include them?
3. Regarding the hyperparameter K (number of factors), could you provide guidance or a quantitative method for determining an appropriate value when working with our own datasets? In the tutorial, K was chosen as 20, but it would be helpful to understand the reasoning behind this choice. Also, how sensitive are the results to changes in K? For example, if we chose K as 21 instead, would the results be significantly affected?

Thank you for providing such a powerful tool for perturbation analysis.

Best,
Huajun

[LifanLiang]

Hi Huajun,

Thank you for your feedback. And I apologize for the very late response.

For now, I can answer the second issue you raised. LUHMES CROP-seq analysis has a dedicated code repository. Functions you mentioned were located at: https://github.com/xinhe-lab/GSFA_paper/blob/master/R/plotting_functions.R. Please download the script and import those functions similar to:

source("../R/plotting_functions.R")

before running the tutorial.

Best,
Lifan

[xinhe2]

Hi Huajun,

Somehow we missed the message you sent a long time ago. Sorry about this. Next time, you could also just email me directly (xinhe@uchicago.edu). About your questions, I'll answer 1) and 3):

1. You said you could't reproduce the results. What problems do you have? Does the program converge?

2. Our Supplementary Notes described how to select K, see Page 17: https://static-content.springer.com/esm/art%3A10.1038%2Fs41592-023-02017-4/MediaObjects/41592_2023_2017_MOESM1_ESM.pdf

Best,
Xin