Validate the clusters from degPatterns

[lgallegovillar]

Dear Lorena,

I am testing degPatterns() to identify gene clusters exhibiting particular patterns across samples of RNA seq data. I would like to validate the clusters identified by computing silhouette. The results from degPatterns identified 9 clusters. I extracted hr, cutree, and looked at the size of the clusters. However, the sizes of the clusters are different from df. Shouldn´t be equal?
I tried to compute silhouette and the results showed a low ave.sil.width. How would you compute silhouette? I appreciate very much any suggestion.

results_degpatterns <- degPatterns(mydatamatrix, metadata= colData,
time= "Time", plot=T, col="Treatment",
reduce=TRUE, minc = 1)
df <- results_degpatterns$df
table(df$cluster)
1 2 3 4 5 6 7 8 9
1516 1616 68 108 11 135 23 5 6

hr <- results_degpatterns$hr
h_hr = hr$dc
grp <- cutree(as.hclust(hr), h=h_hr)
table(grp)
grp
1 2 3 4 5 6 7 8 9
1581 1645 73 116 13 145 26 6 6

m= 1-cor(t(mydatamatrix), method="kendal")
d= as.dist(m^2)
sihr <- silhouette(cutree(as.hclust(hr), h=h_hr),dist=d)
fviz_silhouette(sihr) #library(factoextra)

cluster size ave.sil.width
1 1 1581 0.15
2 2 1645 0.08
3 3 73 0.19
4 4 116 0.33
5 5 13 0.22
6 6 145 0.24
7 7 26 0.18
8 8 6 0.15
9 9 6 0.12

Thank you in advance,
Lorena Gallego

[lpantano]

Hi,
sorry for the delay. Busy week. Can you try with reduce=FALSE, after clustering, there is a function that will reduce outliers over the mean of the cluster expression pattern. For now, that is the one I can think of. Let me know.

[lgallegovillar]

Thanks a lot, it worked.