FUNKI is a multi-omic functional integration and analysis platform. It provides a standardised pipeline to process and perform functional analysis on transcriptomic, proteomic, phosphoproteomic and metabolomic datasets. The analysis can be performed both on a single type of omic data and on multi-omic dataset by integrating them in supervised and unsupervised manners.
FUNKI is accessible through https://funki.shinyapps.io/funki/ or to run locally.
| :warning: | Online version does NOT support CARNIVAL/COSMOS to be run with cplex or cbc. This software is licenced-based, so the LOCAL version of FUNKI is advided (previous installation of the selected software) | |—————|:————————|
To run FUNKI locally, there are different options to fit all possible users:
In an R session, run the line below to launch FUNKI:
shiny::runGitHub(repo = "ShinyFUNKI", username = "saezlab", subdir = "FUNKI")
| :point_up: | Remember to have all required packages installed beforehand! | |—————|:————————|
renv.lock lockfile (
renv packaged) has recorded the state of this project’s private library.
It can be used to restore the state of that library as required by calling
In an R session, run the line below to launch FUNKI once this repo has been downloaded/cloned:
As in case 1, all required packages must be installed beforehand.
renv.lock lockfile can be used to resore the library by calling
The docker file is provided to create a docker imagine.
To build the docker image:
FUNKI/renv.lockin the directory the imagine is going to be created.
Dockerfileis located), run:
docker build -t funki .
This process takes some time (~ 2400s).
To create a container just run:
docker run --rm -p 3838:3838 funki
And there it is, running on
| :point_up: | Remember to install cplex or cbc to use CARNIVAL/COSMOS with these softwares! | |—————|:————————|
DoRothEA (Discriminant Regulon Expression Analysis) is a resource that links transcription factors (TFs) with their downstream targets ( Garcia-Alonso et al., 2018, 2019 ). The unity of a TF and its targets is called regulon. The regulons are built from four different strategies: (i) manually curated interaction repositories, (ii) interactions derived from ChIP-seq binding data, (iii) in silico predictions of TF binding on gene promoters, and (iv) reverse-engineered regulons from gene expression datasets. The TFs activities are computed from gene expression by performing an enrichment analysis ( Alvarez et al., 2016 ), where the regulons are the underlying gene-sets. The users can select the confidence level (A to E based on the type of the supporting evidence of given interactions) for each regulon, as well as their minimum size and the method to perform the enrichment analysis. The organism selection, human or mice, is selected when the data are uploaded ( Holland, Szalai, et al., 2020 ). This method can also be used with single-cell data ( Holland, Tanevski, et al., 2020 ). The computation yields a matrix with the normalised enrichment scores for each TF across all samples. This result is then visualised in the form of a heatmap, barplots and a network showing a TF with all its targets.
PROGENy (Pathway RespOnsive GENes) is a footprint method developed to infer pathway activity from gene expression data (Schubert et al., 2018). The scores are calculated using a linear models with weights based on consensus gene signatures obtained from publicly available perturbation experiments. This method can be used for either bulk or single-cell data Holland, Tanevski, et al., 2020 from human or mouse ( Holland, Szalai, et al., 2020 ). They can also select the number of top genes from the signatures according to their individual significance. This last parameter is particularly important for the single-cell data as it counteracts the typical low gene coverage of this data type. PROGENy returns a matrix of pathway activity scores across all samples. This result is then visualized as a heatmap, barplot and density-scatter plot.
KinAct is a resource linking kinases to phosphorylation sites ( Wirbel et al., 2018 ). It is fully integrated into Omnipath due to the addition of kinase-substrate interaction resources ( Türei et al., 2021, 2016 ). Kinase activity estimation is performed using the same algorithm as DoRothEA to estimate activity scores ( Alvarez et al., 2016 ). Instead of TF-target interactions, KinAct uses collections of kinase-substrate interactions and phosphoproteomic data instead of transcriptomic data. The users can select the minimum size of each regulon, as well as the method that VIPER will use to perform the analysis. The result is a matrix of normalised enrichment scores for each phosphosite across all samples. This result is then visualised in the form of a heatmap, barplots and a network showing a kinase with the targeted phosphosites.
CARNIVAL (CAusal Reasoning for Network identification using Integer VALue programming) reconstructs signalling networks from downstream TF activities by finding the upstream regulators ( Liu et al., 2019 ). Given a directed prior-knowledge network (PKN) of protein-protein interactions, which can also be signed, CARNIVAL identifies a subnetwork that explains the activities of transcription factors through potential perturbed intermediate genes. The PKN can be provided by the user, or imported from Omnipath directly within FUNKI ( Türei et al., 2016, 2016 ). Similarly, the user can directly upload the activity of transcription factors, but those can also be calculated using DoRothEA from the expression data. When large initial networks are used, mapping key nodes with values is advised. Thus, the user can upload them or get advantage of the PROGENy scores for this task. As the previous methods, it can run on mouse or human samples. When using mouse data, the network must always be provided. CARNIVAL produces a set of networks that can be directly visualised. A pathway enrichment analysis can be run over the results. Then, these data can be visualised on bar and volcano plots.
COSMOS is a tool to integrate multiomic data with a prior knowledge network spanning signaling, gene regulation and metabolism ( Dugourd et al. 2021 ). It uses the ILP formulation of CARNIVAL to connect two sets of upstream and downstream molecular features (e.i. kinase activities, TF activities, deregulated metabolites, enzyme fluxes, genetic or drug perturbations, etc…) with a signed directed transomic network. This resulting network is essentially a set of coherent mechanistic hypotheses that can explain how the measured deregulation may explain each other. Subsets of this network centered on user-defined nodes can be viewed in the shiny app. The network can also be downloaded as a pair of sif and attribute csv files. These files can be imported in tools such as cytoscape to visualise the full network.
DoRothEA, PROGENy, CARNIVAL and COSMOS can be either applied on mouse or human data. Independently of the omics technology, they all require a gene expression object with HGNC/MGI symbols in rows and samples in columns. KinAct can only be applied to human data, and it requires a phosphoproteomic object with HGNC symbols and the phosphorilated site in rows, and samples in columns.
DoRothEA, KinAct and PROGENy can compute the respective activities for multiple contrast/samples in a single run. However, CARNIVAL and COSMOS are running on only one sample. For this analysis, the time to find an optimal solution is set to 1h.
Details of each of the required parameters can be found by clicking in the ? symbol. Click on each of the logos to see the parameters required for each computation.
Table of TF-activities.
Table of PROGENy-scores.
Table of kinase activities.
Table of the geneset/pathway enritchment analysis.
Table of the geneset/pathway enritchment analysis.