Direct visualization of DEGs

Two functions in pathlinkR can be applied to a table of DEGs, enabling simple visualization and comparison of results; both can take standard data objects produced by DESeq2 [7] and edgeR [8] as input. The first is “eruption”, which creates a Volcano plot mapping fold change values on the x axis and statistical significance (negative log of the p value) on the y axis. Multiple refinements allow the resulting plot to be tweaked as desired, such as changing colours, adding a summary subtitle, and plotting fold changes as log-transformed or not. Selected genes can be highlighted by providing a list of gene names/IDs, or even the name of a pathway from which constituent genes are automatically highlighted. Pathways accessed by pathlinkR are from Reactome [9], a curated and peer-reviewed database of >2,600 pathways in Homo sapiens, as well as thousands of pathways from other model organisms, organized hierarchically based on cell-specific processes.

The second method for DEG visualization is “plotFoldChange,” which creates a heatmap with genes as rows and DE comparisons as the columns. This allows comparison of how the same set of genes behave in different comparisons of interest. The genes included in the rows can also be selected by providing a list of gene names/IDs, or the name of a Reactome pathway.

Network analysis

A DEG table can be used as input to “ppiBuildNetwork” to create a PPI network, using curated molecular human interaction data from InnateDB [10] (previously not available within the R framework), an IMEX consortium database, and leveraging functions from the igraph [11] and tidygraph [12] packages. Options for network order including zero, first, or minimum are available to suit the number of genes used as a seed for the network. A zero order network contains only genes or nodes that are input by the user, referred to as seed nodes—no additional genes are added via interactions. First order networks include genes which interact with the seed nodes, called interactors, growing and expanding the network using the available data. A minimum order network starts as first order, but any interactors with only a single connection to a seed node are removed, trimming down the overall network. Once built, this network can be visualized directly within R via “ppiPlotNetwork,” which can easily map a variable of interest (e.g. fold change directionality, statistical significance, or node degree) to node colour, and highlight and label key “hub” nodes which are integral in binding the network together. PPI networks can be easily tested for enriched pathways using “ppiEnrichNetwork” (it uses the same methodology as the main pathway enrichment function; see the next section for details) and a sub-network consisting of nodes present in a pathway of interest can be plotted by using “ppiExtractSubnetwork,” followed by the same “ppiPlotNetwork” function to create an updated graph highlighting nodes from the pathway. These methods are specifically designed to work with the results from human RNA-Seq experiments, providing previously unavailable network functionality in the context of analyzing DEGs.

Pathway enrichment

The “pathwayEnrichment” function can be used on a list of one or more DEG tables, testing each for enriched pathways, and returning a single table to facilitate simple comparison of results, making pathlinkR inherently scalable to complex experiments. Three methodologies for enrichment are included: The first is simple over-representation analysis, utilized by the popular ReactomePA [4] package, which treats genes in a pathway individually when calculating the statistical significance of enrichment. The second is gene-pair over-representation analysis, utilized by the Sigora package [13], which treats genes in a pathway as unique pairs to provide less redundant pathway enrichment results. Both options are available when analyzing the Reactome [9] or KEGG [14] human pathway databases, while a second option uses simple over-representation analysis with the Molecular Signatures Database (MSigDB) Hallmark Collection [15] for humans, which are gene sets that represent “specific, well-defined biological states or processes with coherent expression”. Finally, users may also perform gene set enrichment analysis (GSEA) built on the fgsea [16] package, on either Reactome pathways or MSigDB Hallmark terms. These methods have been provided to enable users to have more freedom in determining which methodology and database suits their analysis. Unlike most pathway enrichment tools, for “pathwayEnrichment” the input DEGs from a single comparison can be treated as a complete set, or split based on the directionality of each gene’s fold change, resulting in pathways assigned as “upregulated” (i.e. enriched by upregulated genes, meaning the pathway overall can be thought of as being activated) or “downregulated.” The output format is shared for all available methods and databases, making it simple to incorporate any of these tests into a larger pipeline.

The table of results from “pathwayEnrichment” can be visualized with “pathwayPlots” to create a compelling figure which summarizes and compares the enrichment results across multiple DEG comparisons, scaling to experiments with multiple conditions of interest. Each comparison (e.g. Treatment A vs. Control and Treatment B vs. Control) is represented as a column on the x axis, while significant pathways are placed alphabetically as the rows on the y axis. A triangle indicates a given pathway was significantly enriched in one comparison, while its direction (pointing up or down) denotes if the pathway was derived from up- or down-regulated genes; or if there is no separation by directionality, circles are used instead. When using any of the methods or databases available, the pathways on the y axis can be grouped according to higher level categories, such as “Immune System” or “Metabolism,” to increase interpretability. A number of options can be used to tweak and customize the final output, such as only showing pathways from certain categories, or adding the number of DEGs below each comparisons’ name on the x axis.

Pathway-based networks

The pathway enrichment results from a single DE comparison can be visualized as a network using the “pathnet” family of functions. Each pathway becomes a node in a network, with interactions or edges defined by the number of shared genes between two pathways. Since many genes can have multiple functions and play a role in different pathways, this type of analysis provides an alternative method of visualization that groups seemingly unrelated pathways based on their shared constituent genes, identifying similarities that may otherwise be overlooked. To determine if two pathways are connected, a number of standard distance measures are supported (e.g. Jaccard, Euclidean, and Manhattan), and a threshold is set to define a pair of pathways as being interconnected. By default, the Jaccard Index is used to measure the distance of pathway pairs, and a threshold of 0.8 is used to define a pathway pair as connected. Communities or subgraphs of nodes are not strictly defined, but rather arise organically based on the methods and options set by the user. Two options are provided for visualizing these pathway-based networks within R. The first produces a static plot of the network using the ggraph [17] library; the second uses visNetwork [18] and is interactive, allowing nodes to be selected, dragged, and grouped together, or categories of nodes (e.g. “Immune System”) can be highlighted—all within the R interface. This dynamic plot can be saved as a standard image file for inclusion in a summary presentation or publication.