Mem Plot Concepts

library(multienrichjam)
#> 

Where is the Multi Enrichment Map?

In practice, we rarely find benefit from the multi-enrichment map. The network shows pathways connected to pathways based upon Jaccard overlap of the genes involved in enrichment of each pathway.

This network view was often disorganized, not clearly clustered, and lacked ability to see which genes drive the overlaps between pathways.

Perhaps it works best for the highly structured Gene Ontology (GO), but in our hands GO was just not insightful for the broad range of experiments we were analyzing.

Concept networks (Cnet)

We gravitated toward the Cnet plot, a clever idea by Dr. Guangchang Yu to visualize pathways connected to genes, where genes also connect naturally to other pathways.

For many of our collaborators, this plot is visually intuitive. It also answers the next question people often have:

“What are the shared genes?”

Our subtle customization is to color genes by enrichment to show which genes are shared or unique across enrichments.

The gene nodes are sorted by color, and optionally by border, to help organize patterns.

Two Cnet paradigms

The evolution of Cnet plots led to two conceptual ideas, both driven by the “too many pathways” problem:

1. Cnet using “exemplar pathways”

   • This option is not comprehensive, as it only shows a subset of pathways.
   • This option will not show every gene involved in enrichment.
   • Its main utility is to produce a clean figure. Ultimately, this is also a core goal of multienrichjam.
   • This option is ideal when there are relevant pathways known to be relevant to the experiment, and when Option 2 (below) is too complex.
   • Option 1 is the path chosen by the ggtangle component of the clusterProfiler suite by Dr. Yu. It displays the top N pathways, with default showCategory=5. In our experience, the “top N” are not always the most representative, nor the most interesting. So we extended multienrichjam accordingly.

2. Cnet using pathway clusters

   • This option is intended to be more comprehensive, less complex than plotting 20 pathways individually, but more complex than “Cnet exemplars” above.
   • This option will show every gene involved in enrichment.
   • This option is the main utility of multienrichjam.
   • Pathways are clustered based upon the genes they contain. The approach is similar to Enrichment Map (Bader lab) use of Jaccard overlap. However, clustering is a rich field with many important techniques not captured by a simple overlap coefficient.

Option 2 has several important benefits

• Reduces redundancy among pathways.

  • When the genes involved in pathway enrichment are identical across several pathways, they naturally cluster together.
  • Over much time, we observed that gene-pathway clustering provides a rich overview of the data involved, and gives insight into the underlying pathways and supporting data.

• Avoids clustering by P-value

  • To be frank, our original approach was to plot the P-value matrix of test versus pathway. As a heatmap, it naturally provides clusters, and it sometimes fortuitously appears biologically relevant.
  • We learned that clustering by P-value is incorrect. Instead, we recommend using the underlying gene content of the pathways. The purpose is to group pathways conceptually, and enrichment P-value is not an indication of biology.

• Provides functional sub-groups for interpretation.

  • Ideally, nearly identical pathways are grouped together and are “easily summarized” by a scientist.
  • More often, pathway clusters contain similar pathways, grouped because they also share “core genes” across these pathways.
  • This is an exciting finding in itself, and turns out to be the major step forward in interpreting pathway enrichment findings. Unfortunately, it doesn’t provide an “easy summary” to use as a label.

Guidance

Frankly, neither approach is always comprehensive. Whether comforting, or not, no analysis is itself comprehensive, especially considering the pathway (gene set) resources are themselves imperfect. Rarely are all genes tested for enrichment also present in the pathways being tested, nor present in statistically significant pathways found.

Cnet plots typically cannot show all functionally relevant genes from an experiment.

• Not all functionally relevant genes are annotated.
• Not all functionally relevant genes are assigned to a known pathway or gene set.
• Not all pathways to which a gene is assigned meet statistical thresholds for enrichment.

The primary purpose of a Cnet plot is to visualize the relationship between pathways and genes*, with the following:

• for pathways that met the filtering criteria as tested.

• for genes that were associated with those pathways.

• for pathways arranged in one of two ways:

  1. Cnet exemplars, representing a subset of pathways which are relevant to the particular research study.
  2. Cnet clusters, representing functional groups of pathways based upon shared genes in those pathways.

The secondary purpose of a Cnet plot is to allow scientists to review the specific genes involved, including:

• the relationship of genes to one or more pathways, and
• the association of genes with one or more enrichment tests.