matchSignatures color might cause misunderstanding

[jmche]

I saw the raw code of matchSignatures and I think the function's color might cause misunderstanding.

1. The distance is cosine distance and It's more different if the cosine distance is close to 1 and more similar if the distance is close to 0
2. The color is red if it's close to 1.
3. People would think the red shows more similar, however from the code It is not.
   I think it's better to show the similarity in the plot instead of the distance

[dami82]

Thanks for the comment. Your point is well taken, I guess that using similarity (defined as 1 - distance) instead of distance in the heatmap makes sense. I agree that it is more intuitive to look for signal instead of absence of signal. I will work on the issue you opened in the next few weeks (sorry, but I really have no bandwidth in the next ~3 weeks).

However, it is inaccurate to say that in its current implementation, my code returns red when the distance is ~1. You may have missed the following line:

p  <- p + ggplot2::scale_fill_gradient2(low = "#f40000", mid = "#ffe9e9", high = "white",
                                         midpoint = nuMid, guide = "colourbar",
                                         name = paste(method, "\ndistance", sep = "")

Briefly, this flips the color palette so that the color is white (high) when the cosine distance is ~1 AND the color is red (#f40000) when the distance is low or close to zero (i.e., high similarity).
You may want to verify this by matching few COSMIC signatures against themselves. Red boxes should highlight the diagonal.

Note: this may seem like a trivial modification of the code, but it may take a little more work than you anticipate cause the cosine distance is the default distance method, but other methods are also supported. Therefore, when I modify the function, I need to make sure that the similarity scores [max.distance - distance_i] are computed accurately, and that the final implementation is reliable independently of the desired distance method.

[jmche]

Yeah, I think I was misunderstanding the color and the value of cosine distance. I didn't see the legend details that's why I thought the red means larger cosine distance and it made me really confused.
According to what you said and I think the plot is OK, however, using similarity will be better to avoid the misunderstanding just like I did.