More than two conditions?

[TheSingleCell]

Hi team, Great work!!! My experimental design has three groups: control, disease, and disease+drug treatment. I wonder if Augur can deal with the dataset that has more than two groups? If so, what would you recommend me to do if I want to prioritize the cell types that are most responsive to the drug treatment? Best, Nelson

[skinnider]

Hi @TheSingleCell - there are a few different ways to analyze this kind of design. Take a look at Supplementary Fig. 3 in the paper for a worked example in the Hrvatin et al., 2018 dataset with three conditions (control, 1h and 4h of light stimulation). Basically, you can either analyze all pairwise comparisons (control-disease, control-drug, disease-drug, or some subset of them that you find most interesting), or you can analyze all three conditions jointly in a multiclass classification framework. Without knowing anything else about your dataset my impression is that pairwise comparisons might be more appropriate, but it's hard to say. You could try running both approaches and comparing the results, as in this figure. Hope this helps.

[BioMedInvestorNL]

Hi @TheSingleCell - there are a few different ways to analyze this kind of design. Take a look at Supplementary Fig. 3 in the paper for a worked example in the Hrvatin et al., 2018 dataset with three conditions (control, 1h and 4h of light stimulation). Basically, you can either analyze all pairwise comparisons (control-disease, control-drug, disease-drug, or some subset of them that you find most interesting), or you can analyze all three conditions jointly in a multiclass classification framework. Without knowing anything else about your dataset my impression is that pairwise comparisons might be more appropriate, but it's hard to say. You could try running both approaches and comparing the results, as in this figure. Hope this helps.

Hi. I see this examples and it looks nice, but I am a bit confused on how to make this operational at the code level. at label_col I can define where my conditions are, but how can I actually inform the type of comparisons I want?

[skinnider]

You would need to subset the dataset that you provide as input to Augur. If you provide a dataset with 3+ labels, it will do multiclass classification automatically. If you provide a dataset with just 2 labels it will do binary classification.