Evaluate impact of Semantic Similiarity choice
19 May, 2021
Contents
Introduction
In the overview (seeutils::vignette("overview", package ="ViSEAGO")), we explained how to use ViSEAGO package.
In this vignette we explain how to explore the effect of the GO semantic similarity algorithms on the tree structure, and the effect of the trees clustering based on the mouse_bioconductor vignette dataset (see utils::vignette("2_mouse_bioconductor", package ="ViSEAGO")).
Data
Vignette build convenience (for less build time and size) need that data were pre-calculated (provided by the package), and that illustrations were not interactive.
1 Clusters-heatmap of GO terms
The GO annotations of genes created and enriched GO terms are combined using ViSEAGO::build_GO_SS. The Semantic Similarity (SS) between enriched GO terms are calculated using ViSEAGO::compute_SS_distances method. We compute all distances methods with Resnik, Lin, Rel, Jiang, and Wang algorithms implemented in the GOSemSim package [1]. The built object myGOs contains all informations of enriched GO terms and the SS distances between them.
Then, a hierarchical clustering method using ViSEAGO::GOterms_heatmap is performed based on each SS distance between the enriched GO terms using the ward.D2 aggregation criteria. Clusters of enriched GO terms are obtained by cutting branches off the dendrogram. Here, we choose a dynamic branch cutting method based on the shape of clusters using dynamicTreeCut [2,3].
# compute Semantic Similarity (SS)
myGOs<-ViSEAGO::compute_SS_distances(
myGOs,
distance=c("Resnik","Lin","Rel","Jiang","Wang")
)- Resnik distance
# GO terms heatmap
Resnik_clusters_wardD2<-ViSEAGO::GOterms_heatmap(
myGOs,
showIC=TRUE,
showGOlabels=TRUE,
GO.tree=list(
tree=list(
distance="Resnik",
aggreg.method="ward.D2"
),
cut=list(
dynamic=list(
deepSplit=2,
minClusterSize =2
)
)
),
samples.tree=NULL
)- Lin distance
# GO terms heatmap
Lin_clusters_wardD2<-ViSEAGO::GOterms_heatmap(
myGOs,
showIC=TRUE,
showGOlabels=TRUE,
GO.tree=list(
tree=list(
distance="Lin",
aggreg.method="ward.D2"
),
cut=list(
dynamic=list(
deepSplit=2,
minClusterSize =2
)
)
),
samples.tree=NULL
)- Rel distance
# GO terms heatmap
Rel_clusters_wardD2<-ViSEAGO::GOterms_heatmap(
myGOs,
showIC=TRUE,
showGOlabels=TRUE,
GO.tree=list(
tree=list(
distance="Rel",
aggreg.method="ward.D2"
),
cut=list(
dynamic=list(
deepSplit=2,
minClusterSize =2
)
)
),
samples.tree=NULL
)- Jiang distance
# GO terms heatmap
Jiang_clusters_wardD2<-ViSEAGO::GOterms_heatmap(
myGOs,
showIC=TRUE,
showGOlabels=TRUE,
GO.tree=list(
tree=list(
distance="Jiang",
aggreg.method="ward.D2"
),
cut=list(
dynamic=list(
deepSplit=2,
minClusterSize =2
)
)
),
samples.tree=NULL
)- Wang distance
2 Trees comparison
2.1 Global trees comparisons
The dendextend package [4], offers a set of functions for extending dendrogram objects in R, letting you visualize and compare trees of hierarchical clusterings (see utils::vignette("introduction", package ="dendextend")). In this package we use dendextend::dendlist and dendextend::cor.dendlist functions in order to calculate a correlation matrix between trees, which is based on the Baker Gamma and cophenetic correlation as mentioned in dendextend.
The correlation matrix can be visualized with the nice corrplot::corrplot function from corrplot package [5].
# build the list of trees
dend<- dendextend::dendlist(
"Resnik"=slot(Resnik_clusters_wardD2,"dendrograms")$GO,
"Lin"=slot(Lin_clusters_wardD2,"dendrograms")$GO,
"Rel"=slot(Rel_clusters_wardD2,"dendrograms")$GO,
"Jiang"=slot(Jiang_clusters_wardD2,"dendrograms")$GO,
"Wang"=slot(Wang_clusters_wardD2,"dendrograms")$GO
)
# build the trees matrix correlation
dend_cor<-dendextend::cor.dendlist(dend)
As expected, we can easily tells us that GO semantic similarity algorithms based on the Information Content (IC-based) with Resnik, Lin, Rel, and Jiang methods are more similar than the Wang method which in based on the topology of the GO graph structure (Graph-based).
2.2 Paired trees comparison
We can also compare the dendrograms build with, for example, the Resnik and the Wang algorithms using dendextend::dendlist, dendextend::untangle, and dendextend::tanglegram functions.
The quality of the alignment of the two trees can be calculated with dendextend::entanglement (0: good to 1:bad).
# dendrogram list
dl<-dendextend::dendlist(
slot(Resnik_clusters_wardD2,"dendrograms")$GO,
slot(Wang_clusters_wardD2,"dendrograms")$GO
)
# untangle the trees (efficient but very highly time consuming)
tangle<-dendextend::untangle(
dl,
"step2side"
)
# display the entanglement
dendextend::entanglement(tangle) # 0.08362968
# display the tanglegram
dendextend::tanglegram(
tangle,
margin_inner=5,
edge.lwd=1,
lwd = 1,
lab.cex=0.8,
columns_width = c(5,2,5),
common_subtrees_color_lines=FALSE
)
3 Clusters comparison
Another possibility concerns the comparison of the dendrograms clusters.
3.1 Multiple clusters comparison
We can also explore the GO terms assignation between clusters according the used parameters with ViSEAGO::clusters_cor and plot the results with corrplot::corrplot using corrplot package.
# clusters to compare
clusters=list(
Resnik="Resnik_clusters_wardD2",
Lin="Lin_clusters_wardD2",
Rel="Rel_clusters_wardD2",
Jiang="Jiang_clusters_wardD2",
Wang="Wang_clusters_wardD2"
)
# global dendrogram partition correlation
clust_cor<-ViSEAGO::clusters_cor(
clusters,
method="adjusted.rand"
)# global dendrogram partition correlation
corrplot::corrplot(
clust_cor,
"pie",
"lower",
is.corr=FALSEALSE,
cl.lim=c(0,1)
)
As expected, same as in the global trees comparison, we can easily tells us that GO semantic similarity algorithms based on the Information Content (IC-based) with Resnik, Lin, Rel, and Jiang methods are more similar than the Wang method which in based on the topology of the GO graph structure (Graph-based).
3.2 Paired trees comparison
We can also explore in details the GO terms assignation between clusters according the used parameters with ViSEAGO::compare_clusters.
NB: For this vignette, this illustration is not interactive.
4 Conclusion
ViSEAGO package provides convenient methods to explore the effect of the GO semantic similarity algorithms on the tree structure, and the effect of the trees clustering playing a key role to ensuring functional coherence.
References
1. Yu G, Li F, Qin Y, Bo X, Wu Y, Wang S. GOSemSim: an R package for measuring semantic similarity among GO terms and gene products. Bioinformatics. 2010;26:976–8.
2. Langfelder P, Zhang B, Horvath S. Defining clusters from a hierarchical cluster tree: the Dynamic Tree Cut package for R. Bioinformatics. 2008;24:719–20.
3. Langfelder P, Zhang B, Steve Horvath. DynamicTreeCut: Methods for detection of clusters in hierarchical clustering dendrograms [Internet]. 2016. Available from: https://CRAN.R-project.org/package=dynamicTreeCut
4. Galili T. Dendextend: An r package for visualizing, adjusting, and comparing trees of hierarchical clustering. Bioinformatics [Internet]. 2015; Available from: http://bioinformatics.oxfordjournals.org/content/31/22/3718
5. Wei T, Simko V. Corrplot: Visualization of a correlation matrix [Internet]. 2016. Available from: https://CRAN.R-project.org/package=corrplot