search

ZJUFanLab / scRank

A computational method to rank and infer drug-responsive cell population towards in-silico drug perturbation using a target-perturbed gene regulatory network (tpGRN) for single-cell transcriptomic data
GNU General Public License v3.0
33 stars 3 forks source link

perb_score in the output #8

Open sunnysunnygood opened 3 months ago

sunnysunnygood commented 3 months ago

Dear authors,

Thanks for your amazing tools of scRank!

May I have your guidance that if the results should mainly focus on the rank or the perb_score?

_

As when trying the example in the tutorial, the results shows as below, the resistant cell type rank to be 1st, while the perb_score is lower than the sensitive cell types, which rank to be 2nd.

_

Besides, is there an threshold for the perb_score, 1.01e-6 seems quite small....

Thank you very much for your guidance! Appreciate!

Slot "cell_type_rank":
            perb_score rank top_rank% fill method
resistant 1.101223e-06    1         0    0 scRank
sensitive 2.566986e-06    2         1    1 scRank
Lee0498 commented 1 month ago

Hi, thanks for your feedback.

  1. The rank column is an intermediate variable that reflects the sample ranks of perb_score and can be ignored. The actual ranking of cell types is represented by the top_rank% column.

  2. The small perb_score is due to the nature of the adjacency matrix in the manifold alignment analysis, where values range from 0 to 1, and involve the computation of smallest eigenvalue of the Laplacian matrix.

  3. When interpreting results, it should focus on the top_rank% of cell types rather than the absolute values of perb_score, as there isn't a fixed threshold suitable for every case.