Related to the more robust quantile normalization introduced in 0.12.0. Failing cases are:

• Outliers (large values or large differences in conditions) in the '--signal' sampled footprint scores skews the mean_array_quantiles within quantile_normalization. These outliers have to be handled more robustly.
• Cases where large footprint scores are normalized to 0. Maybe due to ArrayNorm normalization using self.value_max? Or due to the fit in mean_array_quantiles pushing the normalization < 0 -> 0 values?

I am working on fixing this for a 0.12.1-bugfix release.