Opposite signs in lfc and did_lfc from find_de_neighborhood

[chibatoshikaze]

Hi.

Thank you for providing LEMUR packages --- I really appreciate the work you’ve done.

Could you provide the intuitive description when the sign of lfc and did_lfc are opposite sign?

While applying LEMUR to the Human Lung Cell Atlas, I noticed that find_de_neighborhood sometimes returns genes where the lfc and did_lfc values have opposite signs, even when both are statistically significant.

According to Figure 4c in the LEMUR paper, I had expected that a downregulated gene in the neighborhood (i.e., lfc < 0) would typically also show a negative did_lfc, since the difference in effect between the neighborhood and other cells would likely preserve the same sign (i.e., $\Delta_{nbh}$ − $\Delta_{other}$ < 0).

Revisiting your glioblastoma notebook, I also found a similar case: LMO2 has lfc < 0 and did_lfc > 0.

• https://htmlpreview.github.io/?https://github.com/const-ae/lemur-Paper/blob/master/notebooks/glioblastoma_analysis.html

  filter(nei_pan_tumor_small, name == "ENSG00000135363")
  ##              name neighborhood n_cells sel_statistic         pval     adj_pval
  ## 1 ENSG00000135363 cell_4, ....   12598     -47.59512 5.317072e-06 0.0008575923
  ##   f_statistic df1      df2       lfc     did_pval did_adj_pval  did_lfc
  ## 1     43.0769   1 16.67347 -1.819303 7.439143e-05    0.7439143 1.485106

Could you please comment on the intuition behind such cases?
Does this reflect stronger or even opposite regulation in the complement of the neighborhood?

Thanks again for your excellent work.

Best regards.