*Bounty: 50*

*Bounty: 50*

I am using an approach to decompose sigmoidal signals in R. Briefly, signals are decomposed into a subset of components and then a custom value of similarity is computed among samples defined by 2 factors (M = MRS and BSG and S = PU1, WCFS1 and H46), each sample has 2 replicates. I expect a grouping by M or S.

I tried to use this similarity measurement as a distance matrix to perform clustering, but the results make literally no sense. However, when I use this similarity matrix as an input for a correlation it gave nice results for different data inputs. I have never read about using a similarity matrix as an input for correlations. Is it statistically correct to do so?

```
hclust(as.dist(M1), method = "ward.D2")
cor(M1, method = "kendall"), type="upper", order="hclust", col=brewer.pal(n=8, name="RdYlBu")
```