Should we aggregate on and off scores or keep them separetly #67
Description
I had a discussion with @wli51 about whether it’s necessary to reduce both the "on" and "off" scores into a single metric. We agreed that doing so might actually lose valuable information. The "on" and "off" signature scores provide interpretable insights that allow someone to better understand how a compound interacts with a cell.
For example, if we were to plot the "on" and "off" signature scores on a quadrant scatter plot, it would give us a clearer picture of how the treatment interacts with the drug.
However, we still face the issue that EMD scores range from 0 to infinity, which can make plot axes difficult to interpret. More importantly, we currently lack a clear interpretation of these scores: for example, what does a difference in magnitude actually represent? How should an EMD (distance) score be contextualized, say, on a 0–100 scale? If such a definition were possible, how could it be incorporated into a normalization function to map scores to a 0–1 range? Finally, what should the shape of that normalization be - for instance, should penalties increase linearly or exponentially as distances grow?