Noise parameters in publication

[fkgruber]

Hi I'm trying to replicate the results in figure 4 M and O but it is not clear in the paper what were the parameters that were used for adding noise and simulate in general. Specifically, what were the values of

• noise_params,
• decays,
• sampling_state,
• noise_type
• outlier_prob
• outlier mean
• outlier scale
• library size effects mean
• library size effects scale
• dropout shape
• dropout percentile

Also when doing the GRN reconstruction of the noisy data did you apply standard preprocessing like normalizations transformations etc? So far the only things that are mentioned is whether imputation was used or not.

thanks FKG

[fkgruber]

I see that the supplement mentions that noise_param=1, decays=0.8, and noise_type='dpd'. However, none of the other parameters are specified. My understanding was that the parameters were set so that the level of noise was comparable to the real data.

Can you provide what parameters values were actually used in paper results? Alternatively, is the noisy data available? The repository only seems to have denoised data.

[rainx0r]

Don't mean to necropost but in order to save anyone still looking at finding the parameters the trouble of going on a scavenger hunt:

The technical noise settings can be found in Supplementary Table S1 in the paper's Supplementary Material. The published version contains noise settings for DS1 to DS8, DS13, DS14. The bioRxiv pre-print version contains settings for DS1 to DS6 which seem to be different to the DS1-6 parameters in the published versions, but I assume the latter are the ones used in the paper results. The parameters documented in that table are outlier_prob, outlier_mean, outlier_scale, librarysize_mean, librarysize_scale, dropouts_k and dropouts_q and low quality threshold, which are the ones that seem to be missing from the Method Details Appendix.