FELLA: an R package to enrich metabolomics data.

BACKGROUND
Pathway enrichment techniques are useful for understanding experimental metabolomics data. Their purpose is to give context to the affected metabolites in terms of the prior knowledge contained in metabolic pathways. However, the interpretation of a prioritized pathway list is still challenging, as pathways show overlap and cross talk effects.


RESULTS
We introduce FELLA, an R package to perform a network-based enrichment of a list of affected metabolites. FELLA builds a hierarchical representation of an organism biochemistry from the Kyoto Encyclopedia of Genes and Genomes (KEGG), containing pathways, modules, enzymes, reactions and metabolites. In addition to providing a list of pathways, FELLA reports intermediate entities (modules, enzymes, reactions) that link the input metabolites to them. This sheds light on pathway cross talk and potential enzymes or metabolites as targets for the condition under study. FELLA has been applied to six public datasets -three from Homo sapiens, two from Danio rerio and one from Mus musculus- and has reproduced findings from the original studies and from independent literature.


CONCLUSIONS
The R package FELLA offers an innovative enrichment concept starting from a list of metabolites, based on a knowledge graph representation of the KEGG database that focuses on interpretability. Besides reporting a list of pathways, FELLA suggests intermediate entities that are of interest per se. Its usefulness has been shown at several molecular levels on six public datasets, including human and animal models. The user can run the enrichment analysis through a simple interactive graphical interface or programmatically. FELLA is publicly available in Bioconductor under the GPL-3 license.


Introduction
This vignette shows the utility of the FELLA package, which is based in a statistically normalised diffusion process (Picart-Armada et al. 2017), on non-human organisms. In particular, we will work on a multi-omic Mus musculus study. The original study (Gogiashvili et al. 2017) presents a mouse model of the non-alcoholic fatty liver disease (NAFLD). Metabolites in liver tissue from leptin-deficient ob/ob mice and wild type mice were compared using Nuclear Magnetic Resonance (NMR). Afterwards, quantitative real-time polymerase chain reaction (qRT-PCR) helped identify changes at the gene expression level. Finally, biological mechanisms behind NAFLD were elucidated by leveraging the data from both omics.

1.1
Building the database The first step is to build the FELLA.DATA object for the mmu organism from the KEGG database (Kanehisa et al. 2016). We load the FELLA.DATA object and two mappings (from gene symbol to entrez identifiers, and from enzyme EC numbers to their annotated entrez genes).

Note on reproducibility
We want to emphasise that FELLA builds its FELLA.DATA object using the most recent version of the KEGG database. KEGG is frequently updated and therefore small changes can take place in the knowledge graph between different releases. The discussion on our findings was written at the date specified in the vignette header and using the KEGG release in the Reproducibility section.
2 Enrichment analysis 2.1 Defining the input and running the enrichment Table 2 from the main body in (Gogiashvili et al. 2017) contains six metabolites that show significant changes between the experimental classes by a univariate test followed by multiple test correction. These are the start of our enrichment analysis: The highlighted subgraph with the default parameters has the following appeareance, with large connected components that involve the metabolites in the input: We will also extract all the p-scores and the suggested sub-network for further analysis:  Table 2 The authors find 5 extra metabolites in Table 2 that are significant at p < 0.05 but do not appear after thresholding the false discovery rate at 5%. Such metabolites, highlighted in italics but without an asterisk, are also relevant and play a role in their discussion. We will examine how FELLA prioritises such metabolites: When checking if any of these metabolites are found in the reported sub-network, we find that C01026 is already reported: Abbreviated as DMG in their study, N,N-Dimethylglycine is a cornerstone of their findings. It is reported in Figure 6a as part of the folate-independent remethylation to explain the metabolic changes observed in the ob/ob mice. DMG is also mentioned in the conclusions as part of one of the most prominent alterations found in the study: a reduced conversion of betaine to DMG.

From Figure 6a
Figure 6a contains the metabolic context of the observed alterations, with processes such as transsulfuration and folate-dependent remethylation. These were identified with the help of gene expression analysis. We will now check for coincidences between the metabolites in Figure 6a, excluding choline and betaine for being in the input and DMG since it was already discussed.  The test is indeed significant -despite FELLA does not directly report such metabolites, its metabolite ranking supports the claims by the authors.

Examining the genes 2.3.1 Cbs
The authors complement the metabolomic profilings with a differential gene expression study. One of the main findings is a change of Cbs expression levels. To link Cbs to the enrichment from FELLA, we will first map it to its EC number As shown in Figure 6a, Cbs is not directly linked to the metabolites found through NMR, and nor the reaction neither the enzyme are suggested by FELLA: It's not surprising that none of them is directly reported, because none of the metabolites participating in the reaction is found in the input. The main evidence for finding Cbs is gene expression, and our approach gives indirect hints of this connection.

Bhmt
The alteration of Bhmt activity is related to the downregulation of Cbs. Despite not finding evidence of change in Bhmt expression, the authors argue that its inhibition would explain the increased betaine-to-DMG ratio in ob/ob mice. Such claim is also backed up by prior studies. To find out the role of Cbs in our analysis, we will again map it to its EC number, 2. This time, FELLA not only reports it, but also its associated reaction R02821 (represented by an arrow in Figure 6a) and both of its metabolites. While betaine was already an input metabolite, DMG was a novel finding as discussed earlier

Slc22a5
The decrease of Bhmt activity is later connected to the upregulation of Slc22a5, also proved within the original study. However, Slc22a5 does not map to any EC number and therefore it cannot be found through FELLA: entrez.slc22a5 %in% names(entrez2ec)

## [1] FALSE
As a matter of fact, the only connection that can be found from KEGG is the role of Slc22a5 in the Choline metabolism in cancer pathway.

Genes from Figure 3
We also examined if genes from Table 3 were reachable in our analysis. These five literaturederived genes were experimentally confirmed to show gene expression changes, in order to prove that RNA extracted after the metabolomic profiling was still reliable for further transcriptomic analyses. However, only Scd2 maps to an enzymatic family: The fact that only one gene mapped to an EC number hinders the potential findings using FELLA, and is probably the main reason why FELLA missed Scd2. In addition, FELLA defines a knowledge model that offers simplicity and interpretability, at the cost of introducing limitations on how sophisticated its findings can be.

Genes from Table S2
In parallel with the original study, and cited within its main body, gene array expression data was collected (Godoy et al. 2016) and its hits are included in the supplementary Table S2 from (Gogiashvili et al. 2017). These genes include the already discussed Cbs. We will attempt to link the genes marked as significantly changed to our reported sub-network. In contrast with Figure 3, all the genes map to an EC number: These findings suggest that if the annotation database is complete enough, FELLA can provide a meaningful priorisisation of the enzymes surrounding the affected metabolites.

Conclusions
FELLA has been used to give a biological meaning to a list of 6 metabolites extracted from a multi-omic study of a mouse model of NAFLD. It has been able to reproduce some findings at the metabolite and gene expression levels, whereas most of the times missed entities would still present a low ranking compared to their background in the database. The bottom line from our analysis in the present vignette is that FELLA not only works on human studies, but also generalises to animal models. Image of the workspace (for submission): tempfile(pattern = "vignette _ mmu _ ", fileext = ".RData") %T>% message("Saving workspace to ", .) %>% save.image(compress = "xz") ## Saving workspace to /tmp/RtmpQORpuY/vignette _ mmu _ 2e33f36219d.RData