|
a |
|
b/vignettes/netOmics.Rmd |
|
|
1 |
--- |
|
|
2 |
title: "netOmics" |
|
|
3 |
author: |
|
|
4 |
- name: "Antoine Bodein" |
|
|
5 |
affiliation: "CHU de Québec Research Center, Université Laval, Molecular Medicine department, Québec, QC, Canada" |
|
|
6 |
email: "antoine.bodein.1@ulaval.ca" |
|
|
7 |
- name: "Marie-Pier Scott-Boyer" |
|
|
8 |
affiliation: "CHU de Québec Research Center, Université Laval, Molecular Medicine department, Québec, QC, Canada" |
|
|
9 |
- name: "Olivier Perin" |
|
|
10 |
affiliation: "Digital Sciences Department, L’Oréal Advanced Research, Aulnay-sous-bois, France" |
|
|
11 |
- name: "Kim-Anh Lê Cao" |
|
|
12 |
affiliation: "Melbourne Integrative Genomics, School of Mathematics and Statistics, University of Melbourne, Melbourne, VIC, Australia" |
|
|
13 |
- name: "Arnaud Droit" |
|
|
14 |
affiliation: "CHU de Québec Research Center, Université Laval, Molecular Medicine department, Québec, QC, Canada" |
|
|
15 |
package: netOmics |
|
|
16 |
output: |
|
|
17 |
BiocStyle::html_document |
|
|
18 |
|
|
|
19 |
vignette: > |
|
|
20 |
%\VignetteIndexEntry{netOmics} |
|
|
21 |
%\VignetteEngine{knitr::rmarkdown} |
|
|
22 |
%\VignetteEncoding{UTF-8} |
|
|
23 |
|
|
|
24 |
bibliography: ["mybib.bib"] |
|
|
25 |
biblio-style: apalike |
|
|
26 |
link-citations: true |
|
|
27 |
--- |
|
|
28 |
|
|
|
29 |
```{r, echo=FALSE} |
|
|
30 |
knitr::opts_chunk$set(fig.align = "center") |
|
|
31 |
``` |
|
|
32 |
|
|
|
33 |
|
|
|
34 |
The emergence of multi-omics data enabled the development of |
|
|
35 |
integration methods. |
|
|
36 |
|
|
|
37 |
With netOmics, we go beyond integration by introducing an interpretation tool. |
|
|
38 |
netOmics is a package for the creation and exploration of multi-omics networks. |
|
|
39 |
|
|
|
40 |
Depending on the provided dataset, it allows to create inference networks from |
|
|
41 |
expression data but also interaction networks from knowledge databases. |
|
|
42 |
After merging the sub-networks to obtain a global multi-omics network, |
|
|
43 |
we propose network exploration methods using propoagation techniques to perform |
|
|
44 |
functional prediction or identification of molecular mechanisms. |
|
|
45 |
|
|
|
46 |
Furthermore, the package has been developed for longitudinal multi-omics data |
|
|
47 |
and can be used in conjunction with our previously published package timeOmics. |
|
|
48 |
|
|
|
49 |
 |
|
|
50 |
|
|
|
51 |
For more informnation about the method, please check [@bodein2020interpretation] |
|
|
52 |
|
|
|
53 |
In this vignette, we introduced a case study which depict the main steps to |
|
|
54 |
create and explore multi-omics networks from multi-omics time-course data. |
|
|
55 |
|
|
|
56 |
# Requirements |
|
|
57 |
|
|
|
58 |
```{r,eval=FALSE} |
|
|
59 |
# install the package via BioConductor |
|
|
60 |
if (!requireNamespace("BiocManager", quietly = TRUE)) |
|
|
61 |
install.packages("BiocManager") |
|
|
62 |
|
|
|
63 |
BiocManager::install("netOmics") |
|
|
64 |
``` |
|
|
65 |
|
|
|
66 |
```{r,eval=FALSE} |
|
|
67 |
# install the package via github |
|
|
68 |
library(devtools) |
|
|
69 |
install_github("abodein/netOmics") |
|
|
70 |
``` |
|
|
71 |
|
|
|
72 |
|
|
|
73 |
```{r, eval=TRUE, message=FALSE} |
|
|
74 |
# load the package |
|
|
75 |
library(netOmics) |
|
|
76 |
``` |
|
|
77 |
|
|
|
78 |
|
|
|
79 |
```{r, eval=TRUE, message=FALSE} |
|
|
80 |
# usefull packages to build this vignette |
|
|
81 |
library(timeOmics) |
|
|
82 |
library(tidyverse) |
|
|
83 |
library(igraph) |
|
|
84 |
``` |
|
|
85 |
|
|
|
86 |
# Case Study: Human Microbiome Project T2D |
|
|
87 |
|
|
|
88 |
The package will be illustrated on longitudinal MO dataset to study |
|
|
89 |
the seasonality of MO expression in patients with diabetes [@sailani2020deep]. |
|
|
90 |
|
|
|
91 |
The data used in this vignette is a subset of the data available at: |
|
|
92 |
https://github.com/aametwally/ipop_seasonal |
|
|
93 |
|
|
|
94 |
We focused on a single individual with 7 timepoints. |
|
|
95 |
6 different omics were sampled |
|
|
96 |
(RNA, proteins, cytokines, gut microbiome, metabolites and clinical variables). |
|
|
97 |
|
|
|
98 |
```{r load_data} |
|
|
99 |
# load data |
|
|
100 |
data("hmp_T2D") |
|
|
101 |
``` |
|
|
102 |
|
|
|
103 |
|
|
|
104 |
# (optional: *timeOmics* analysis) |
|
|
105 |
|
|
|
106 |
The first step of the analysis is the preprocessing and longitudinal clustering. |
|
|
107 |
This step is carried out with timeOmics and should be reserved for longitudinal |
|
|
108 |
data. |
|
|
109 |
|
|
|
110 |
It ensures that the time profiles are classified into groups of similar profiles |
|
|
111 |
so each MO molecule is labbeled with its cluster. |
|
|
112 |
|
|
|
113 |
In addition, timeOmics can identify a multi-omics signature of the clusters. |
|
|
114 |
These molecules can be, for example, the starting points of the propogation |
|
|
115 |
analysis. |
|
|
116 |
|
|
|
117 |
For more informations about *timeOmics*, please |
|
|
118 |
check http://www.bioconductor.org/packages/release/bioc/html/timeOmics.html |
|
|
119 |
|
|
|
120 |
As illustrated in the R chunk below the timeOmics step includes: |
|
|
121 |
|
|
|
122 |
* omic-specific preprocessing and longitudinal fold-change filtering |
|
|
123 |
* modelling of expression profiles |
|
|
124 |
* clustering of MO expression profiles |
|
|
125 |
* signature identification by cluster |
|
|
126 |
|
|
|
127 |
```{r timeOmics_1, eval=FALSE} |
|
|
128 |
# not evaluated in this vignette |
|
|
129 |
|
|
|
130 |
#1 filter fold-change |
|
|
131 |
remove.low.cv <- function(X, cutoff = 0.5){ |
|
|
132 |
# var.coef |
|
|
133 |
cv <- unlist(lapply(as.data.frame(X), |
|
|
134 |
function(x) abs(sd(x, na.rm = TRUE)/mean(x, na.rm= TRUE)))) |
|
|
135 |
return(X[,cv > cutoff]) |
|
|
136 |
} |
|
|
137 |
fc.threshold <- list("RNA"= 1.5, "CLINICAL"=0.2, "GUT"=1.5, "METAB"=1.5, |
|
|
138 |
"PROT" = 1.5, "CYTO" = 1) |
|
|
139 |
|
|
|
140 |
# --> hmp_T2D$raw |
|
|
141 |
data.filter <- imap(raw, ~{remove.low.cv(.x, cutoff = fc.threshold[[.y]])}) |
|
|
142 |
|
|
|
143 |
#2 scale |
|
|
144 |
data <- lapply(data.filter, function(x) log(x+1)) |
|
|
145 |
# --> hmp_T2D$data |
|
|
146 |
|
|
|
147 |
|
|
|
148 |
#3 modelling |
|
|
149 |
lmms.func <- function(X){ |
|
|
150 |
time <- rownames(X) %>% str_split("_") %>% |
|
|
151 |
map_chr(~.x[[2]]) %>% as.numeric() |
|
|
152 |
lmms.output <- lmms::lmmSpline(data = X, time = time, |
|
|
153 |
sampleID = rownames(X), deri = FALSE, |
|
|
154 |
basis = "p-spline", numCores = 4, |
|
|
155 |
keepModels = TRUE) |
|
|
156 |
return(lmms.output) |
|
|
157 |
} |
|
|
158 |
data.modelled <- lapply(data, function(x) lmms.func(x)) |
|
|
159 |
|
|
|
160 |
# 4 clustering |
|
|
161 |
block.res <- block.pls(data.modelled, indY = 1, ncomp = 1) |
|
|
162 |
getCluster.res <- getCluster(block.res) |
|
|
163 |
# --> hmp_T2D$getCluster.res |
|
|
164 |
|
|
|
165 |
|
|
|
166 |
# 5 signature |
|
|
167 |
list.keepX <- list("CLINICAL" = 4, "CYTO" = 3, "GUT" = 10, "METAB" = 3, |
|
|
168 |
"PROT" = 2,"RNA" = 34) |
|
|
169 |
sparse.block.res <- block.spls(data.modelled, indY = 1, ncomp = 1, scale =TRUE, |
|
|
170 |
keepX =list.keepX) |
|
|
171 |
getCluster.sparse.res <- getCluster(sparse.block.res) |
|
|
172 |
# --> hmp_T2D$getCluster.sparse.res |
|
|
173 |
``` |
|
|
174 |
|
|
|
175 |
timeOmics resulted in 2 clusters, labelled `1` and `-1` |
|
|
176 |
|
|
|
177 |
```{r timeOmics_2} |
|
|
178 |
# clustering results |
|
|
179 |
cluster.info <- hmp_T2D$getCluster.res |
|
|
180 |
``` |
|
|
181 |
|
|
|
182 |
# Network building |
|
|
183 |
|
|
|
184 |
Each layer of the network is built sequentially and then assembled |
|
|
185 |
in a second section. |
|
|
186 |
|
|
|
187 |
All the functions in the package can be used on one element or a list of |
|
|
188 |
elements. |
|
|
189 |
In the longitudinal context of the data, kinetic cluster sub-networks are built |
|
|
190 |
plus a global network |
|
|
191 |
(`1`, `-1` and `All`). |
|
|
192 |
|
|
|
193 |
## Inference Network |
|
|
194 |
|
|
|
195 |
Multi-omics network building starts with a first layer of gene. |
|
|
196 |
Currently, the ARACNe algorithm handles the inference but we will include more |
|
|
197 |
algorithms in the future. |
|
|
198 |
|
|
|
199 |
The function `get_grn` return a Gene Regulatory Network from gene expression |
|
|
200 |
data. |
|
|
201 |
Optionally, the user can provide a timeOmics clustering result (`?getCluster`) |
|
|
202 |
to get cluster specific sub-networks. In this case study, this will |
|
|
203 |
automatically build the networks (`1`, `-1` and `All`), as indicated previously. |
|
|
204 |
|
|
|
205 |
The `get_graph_stats` function provides basic graph statistics such as the |
|
|
206 |
number of vertices and edges. |
|
|
207 |
If the vertices have different attributes, it also includes a summary of these. |
|
|
208 |
|
|
|
209 |
```{r graph.rna, warning=FALSE} |
|
|
210 |
cluster.info.RNA <- timeOmics::getCluster(cluster.info, user.block = "RNA") |
|
|
211 |
graph.rna <- get_grn(X = hmp_T2D$data$RNA, cluster = cluster.info.RNA) |
|
|
212 |
|
|
|
213 |
# to get info about the network |
|
|
214 |
get_graph_stats(graph.rna) |
|
|
215 |
``` |
|
|
216 |
|
|
|
217 |
## Interaction from databases |
|
|
218 |
|
|
|
219 |
As for the genes, the second layer is a protein layer (Protein-Protein |
|
|
220 |
Interaction). |
|
|
221 |
This time, no inference is performed. Instead, known interactions are extracted |
|
|
222 |
from a database of interaction (BIOGRID). |
|
|
223 |
|
|
|
224 |
The function `get_interaction_from_database` will fetch the interactions from a |
|
|
225 |
database provided as a `data.frame` (with columns `from` and `to`) or a graph |
|
|
226 |
(`igraph` object). |
|
|
227 |
In addition to the interactions between the indicated molecules, the first |
|
|
228 |
degree neighbors can also be collected (option `user.ego = TRUE`) |
|
|
229 |
|
|
|
230 |
|
|
|
231 |
```{r PROT_graph, warning=FALSE} |
|
|
232 |
# Utility function to get the molecules by cluster |
|
|
233 |
get_list_mol_cluster <- function(cluster.info, user.block){ |
|
|
234 |
require(timeOmics) |
|
|
235 |
tmp <- timeOmics::getCluster(cluster.info, user.block) |
|
|
236 |
res <- tmp %>% split(.$cluster) %>% |
|
|
237 |
lapply(function(x) x$molecule) |
|
|
238 |
res[["All"]] <- tmp$molecule |
|
|
239 |
return(res) |
|
|
240 |
} |
|
|
241 |
|
|
|
242 |
cluster.info.prot <- get_list_mol_cluster(cluster.info, user.block = 'PROT') |
|
|
243 |
graph.prot <- get_interaction_from_database(X = cluster.info.prot, |
|
|
244 |
db = hmp_T2D$interaction.biogrid, |
|
|
245 |
type = "PROT", user.ego = TRUE) |
|
|
246 |
# get_graph_stats(graph.prot) |
|
|
247 |
``` |
|
|
248 |
|
|
|
249 |
In this example, only a subset of the Biogrid database is used |
|
|
250 |
(matching elements). |
|
|
251 |
|
|
|
252 |
## Other inference methods |
|
|
253 |
|
|
|
254 |
Another way to compute networks from expression data is to use other inference |
|
|
255 |
methods. |
|
|
256 |
In the following chunk, we intend to illustrate the use of the SparCC algorithm |
|
|
257 |
[@friedman2012inferring] on the gut data and how it can be integrate into the |
|
|
258 |
pipeline. |
|
|
259 |
(sparcc is not included in this package) |
|
|
260 |
|
|
|
261 |
|
|
|
262 |
```{r GUT_graph, eval = FALSE} |
|
|
263 |
# not evaluated in this vignette |
|
|
264 |
library(SpiecEasi) |
|
|
265 |
|
|
|
266 |
get_sparcc_graph <- function(X, threshold = 0.3){ |
|
|
267 |
res.sparcc <- sparcc(data = X) |
|
|
268 |
sparcc.graph <- abs(res.sparcc$Cor) >= threshold |
|
|
269 |
colnames(sparcc.graph) <- colnames(X) |
|
|
270 |
rownames(sparcc.graph) <- colnames(X) |
|
|
271 |
res.graph <- graph_from_adjacency_matrix(sparcc.graph, |
|
|
272 |
mode = "undirected") %>% simplify |
|
|
273 |
return(res.graph) |
|
|
274 |
} |
|
|
275 |
|
|
|
276 |
gut_list <- get_list_mol_cluster(cluster.info, user.block = 'GUT') |
|
|
277 |
|
|
|
278 |
graph.gut <- list() |
|
|
279 |
graph.gut[["All"]] <- get_sparcc_graph(hmp_T2D$raw$GUT, threshold = 0.3) |
|
|
280 |
graph.gut[["1"]] <- get_sparcc_graph(hmp_T2D$raw$GUT %>% |
|
|
281 |
dplyr::select(gut_list[["1"]]), |
|
|
282 |
threshold = 0.3) |
|
|
283 |
graph.gut[["-1"]] <- get_sparcc_graph(hmp_T2D$raw$GUT %>% |
|
|
284 |
dplyr::select(gut_list[["-1"]]), |
|
|
285 |
threshold = 0.3) |
|
|
286 |
class(graph.gut) <- "list.igraph" |
|
|
287 |
``` |
|
|
288 |
|
|
|
289 |
```{r GUT} |
|
|
290 |
graph.gut <- hmp_T2D$graph.gut |
|
|
291 |
# get_graph_stats(graph.gut) |
|
|
292 |
``` |
|
|
293 |
|
|
|
294 |
## Other examples |
|
|
295 |
|
|
|
296 |
For this case study, we complete this first step of network building with the |
|
|
297 |
missing layers. |
|
|
298 |
|
|
|
299 |
```{r CYTO_graph, warning=FALSE} |
|
|
300 |
# CYTO -> from database (biogrid) |
|
|
301 |
cyto_list = get_list_mol_cluster(cluster.info = cluster.info, |
|
|
302 |
user.block = "CYTO") |
|
|
303 |
graph.cyto <- get_interaction_from_database(X = cyto_list, |
|
|
304 |
db = hmp_T2D$interaction.biogrid, |
|
|
305 |
type = "CYTO", user.ego = TRUE) |
|
|
306 |
# get_graph_stats(graph.cyto) |
|
|
307 |
|
|
|
308 |
# METAB -> inference |
|
|
309 |
cluster.info.metab <- timeOmics::getCluster(X = cluster.info, |
|
|
310 |
user.block = "METAB") |
|
|
311 |
graph.metab <- get_grn(X = hmp_T2D$data$METAB, |
|
|
312 |
cluster = cluster.info.metab) |
|
|
313 |
# get_graph_stats(graph.metab) |
|
|
314 |
|
|
|
315 |
# CLINICAL -> inference |
|
|
316 |
cluster.info.clinical <- timeOmics::getCluster(X = cluster.info, |
|
|
317 |
user.block = 'CLINICAL') |
|
|
318 |
graph.clinical <- get_grn(X = hmp_T2D$data$CLINICAL, |
|
|
319 |
cluster = cluster.info.clinical) |
|
|
320 |
# get_graph_stats(graph.clinical) |
|
|
321 |
``` |
|
|
322 |
|
|
|
323 |
# Layer merging |
|
|
324 |
|
|
|
325 |
We included 2 types of layer merging: |
|
|
326 |
|
|
|
327 |
* *merging with interactions* uses the shared elements between 2 graphs to build |
|
|
328 |
a larger network. |
|
|
329 |
* *merging with correlations* uses the spearman correlation from expression |
|
|
330 |
profiles between 2 layers when any interaction is known. |
|
|
331 |
|
|
|
332 |
|
|
|
333 |
## Merging with interactions |
|
|
334 |
|
|
|
335 |
The function `combine_layers` enables the fusion of different network layers. |
|
|
336 |
It combines the network (or list of network) in `graph1` with the network |
|
|
337 |
(or list of network) in `graph2`, based on the shared vertices between |
|
|
338 |
the networks. |
|
|
339 |
|
|
|
340 |
Additionally, the user can provide an interaction table `interaction.df` |
|
|
341 |
(in the form of a data.frame or igraph object). |
|
|
342 |
|
|
|
343 |
In the following chunk, we sequentially merge RNA, PROT and CYTO layers and uses |
|
|
344 |
the TFome information (TF protein -> Target Gene) to connect these layers. |
|
|
345 |
|
|
|
346 |
```{r, merged_0} |
|
|
347 |
full.graph <- combine_layers(graph1 = graph.rna, graph2 = graph.prot) |
|
|
348 |
full.graph <- combine_layers(graph1 = full.graph, graph2 = graph.cyto) |
|
|
349 |
|
|
|
350 |
full.graph <- combine_layers(graph1 = full.graph, |
|
|
351 |
graph2 = hmp_T2D$interaction.TF) |
|
|
352 |
# get_graph_stats(full.graph) |
|
|
353 |
``` |
|
|
354 |
|
|
|
355 |
## Merging with correlations |
|
|
356 |
|
|
|
357 |
To connect omics layers for which no interaction information is available, |
|
|
358 |
we propose to use a threshold on the correlation between the expression profiles |
|
|
359 |
of two or more omics data. |
|
|
360 |
|
|
|
361 |
The strategy is as follows: we isolate the omics from the data and calculate |
|
|
362 |
the correlations between this omics and the other data. |
|
|
363 |
|
|
|
364 |
```{r merged_1_gut, warning=FALSE} |
|
|
365 |
all_data <- reduce(hmp_T2D$data, cbind) |
|
|
366 |
|
|
|
367 |
# omic = gut |
|
|
368 |
gut_list <- get_list_mol_cluster(cluster.info, user.block = "GUT") |
|
|
369 |
omic_data <- lapply(gut_list, function(x)dplyr::select(hmp_T2D$data$GUT, x)) |
|
|
370 |
|
|
|
371 |
# other data = "RNA", "PROT", "CYTO" |
|
|
372 |
other_data_list <- get_list_mol_cluster(cluster.info, |
|
|
373 |
user.block = c("RNA", "PROT", "CYTO")) |
|
|
374 |
other_data <- lapply(other_data_list, function(x)dplyr::select(all_data, x)) |
|
|
375 |
|
|
|
376 |
# get interaction between gut data and other data |
|
|
377 |
interaction_df_gut <- get_interaction_from_correlation(X = omic_data, |
|
|
378 |
Y = other_data, |
|
|
379 |
threshold = 0.99) |
|
|
380 |
|
|
|
381 |
# and merge with full graph |
|
|
382 |
full.graph <- combine_layers(graph1 = full.graph, |
|
|
383 |
graph2 = hmp_T2D$graph.gut, |
|
|
384 |
interaction.df = interaction_df_gut$All) |
|
|
385 |
``` |
|
|
386 |
|
|
|
387 |
|
|
|
388 |
```{r, merged_2_clinical, warning=FALSE} |
|
|
389 |
# omic = Clinical |
|
|
390 |
clinical_list <- get_list_mol_cluster(cluster.info, user.block = "CLINICAL") |
|
|
391 |
omic_data <- lapply(clinical_list, |
|
|
392 |
function(x)dplyr::select(hmp_T2D$data$CLINICAL, x)) |
|
|
393 |
|
|
|
394 |
# other data = "RNA", "PROT", "CYTO", "GUT" |
|
|
395 |
other_data_list <- get_list_mol_cluster(cluster.info, |
|
|
396 |
user.block = c("RNA", "PROT", |
|
|
397 |
"CYTO", "GUT")) |
|
|
398 |
other_data <- lapply(other_data_list, function(x)dplyr::select(all_data, x)) |
|
|
399 |
|
|
|
400 |
|
|
|
401 |
# get interaction between gut data and other data |
|
|
402 |
interaction_df_clinical <- get_interaction_from_correlation(X = omic_data |
|
|
403 |
, Y = other_data, |
|
|
404 |
threshold = 0.99) |
|
|
405 |
|
|
|
406 |
# and merge with full graph |
|
|
407 |
full.graph <- combine_layers(graph1 = full.graph, |
|
|
408 |
graph2 = hmp_T2D$graph.clinical, |
|
|
409 |
interaction.df = interaction_df_clinical$All) |
|
|
410 |
``` |
|
|
411 |
|
|
|
412 |
|
|
|
413 |
```{r, merged_3_metab, warning=FALSE} |
|
|
414 |
# omic = Metab |
|
|
415 |
metab_list <- get_list_mol_cluster(cluster.info, user.block = "METAB") |
|
|
416 |
omic_data <- lapply(metab_list, function(x)dplyr::select(hmp_T2D$data$METAB, x)) |
|
|
417 |
|
|
|
418 |
# other data = "RNA", "PROT", "CYTO", "GUT", "CLINICAL" |
|
|
419 |
other_data_list <- get_list_mol_cluster(cluster.info, |
|
|
420 |
user.block = c("RNA", "PROT", "CYTO", |
|
|
421 |
"GUT", "CLINICAL")) |
|
|
422 |
other_data <- lapply(other_data_list, function(x)dplyr::select(all_data, x)) |
|
|
423 |
|
|
|
424 |
# get interaction between gut data and other data |
|
|
425 |
interaction_df_metab <- get_interaction_from_correlation(X = omic_data, |
|
|
426 |
Y = other_data, |
|
|
427 |
threshold = 0.99) |
|
|
428 |
|
|
|
429 |
# and merge with full graph |
|
|
430 |
full.graph <- combine_layers(graph1 = full.graph, |
|
|
431 |
graph2 = graph.metab, |
|
|
432 |
interaction.df = interaction_df_metab$All) |
|
|
433 |
``` |
|
|
434 |
|
|
|
435 |
# Addition of supplemental layers |
|
|
436 |
|
|
|
437 |
For the interpretation of the MO integration results, the use of additional |
|
|
438 |
information layers or molecules can be useful to enrich the network. |
|
|
439 |
|
|
|
440 |
## Over Representation Analysis |
|
|
441 |
|
|
|
442 |
ORA is a common step to include knowledge. |
|
|
443 |
The function `get_interaction_from_ORA` perform the ORA analysis from the |
|
|
444 |
desired molecules and return an interaction graph with the enriched terms and |
|
|
445 |
the corresponding molecules. |
|
|
446 |
|
|
|
447 |
Then, the interaction graph with the new vertices can be linked to the network |
|
|
448 |
as illustrated in the previous step. |
|
|
449 |
|
|
|
450 |
Here, ORA was performed with RNA, PROT, and CYTO against the Gene Ontology. |
|
|
451 |
|
|
|
452 |
```{r} |
|
|
453 |
# ORA by cluster/All |
|
|
454 |
mol_ora <- get_list_mol_cluster(cluster.info, |
|
|
455 |
user.block = c("RNA", "PROT", "CYTO")) |
|
|
456 |
|
|
|
457 |
# get ORA interaction graph by cluster |
|
|
458 |
graph.go <- get_interaction_from_ORA(query = mol_ora, |
|
|
459 |
sources = "GO", |
|
|
460 |
organism = "hsapiens", |
|
|
461 |
signif.value = TRUE) |
|
|
462 |
|
|
|
463 |
# merge |
|
|
464 |
full.graph <- combine_layers(graph1 = full.graph, graph2 = graph.go) |
|
|
465 |
``` |
|
|
466 |
|
|
|
467 |
## External knowledge |
|
|
468 |
|
|
|
469 |
Additionally, knowledge from external sources can be included in the network. |
|
|
470 |
|
|
|
471 |
In the following chunk, we performed disease-related gene enrichment analysis |
|
|
472 |
from *medlineRanker* (http://cbdm-01.zdv.uni-mainz.de/~jfontain/cms/?page_id=4). |
|
|
473 |
We converted the results into a data.frame (with the columns `from` and `to`) |
|
|
474 |
and this acted as an interaction database. |
|
|
475 |
|
|
|
476 |
```{r} |
|
|
477 |
# medlineRanker -> database |
|
|
478 |
medlineranker.res.df <- hmp_T2D$medlineranker.res.df %>% |
|
|
479 |
dplyr::select(Disease, symbol) %>% |
|
|
480 |
set_names(c("from", "to")) |
|
|
481 |
|
|
|
482 |
mol_list <- get_list_mol_cluster(cluster.info = cluster.info, |
|
|
483 |
user.block = c("RNA", "PROT", "CYTO")) |
|
|
484 |
graph.medlineranker <- get_interaction_from_database(X = mol_list, |
|
|
485 |
db = medlineranker.res.df, |
|
|
486 |
type = "Disease", |
|
|
487 |
user.ego = TRUE) |
|
|
488 |
# get_graph_stats(graph.medlineranker) |
|
|
489 |
|
|
|
490 |
# merging |
|
|
491 |
full.graph <- combine_layers(graph1 = full.graph, graph2 = graph.medlineranker) |
|
|
492 |
``` |
|
|
493 |
|
|
|
494 |
We complete the MO network preparation with attribute cleaning and addition of |
|
|
495 |
several attributes such as: |
|
|
496 |
|
|
|
497 |
* mode = "core" if the vertex was originally present in the data; "extended" |
|
|
498 |
otherwise |
|
|
499 |
* sparse = TRUE if the vertex was present in kinetic cluster signature; FALSE |
|
|
500 |
otherwise |
|
|
501 |
* type = type of omics ("RNA","PROT","CLINICAL","CYTO","GUT","METAB","GO", |
|
|
502 |
"Disease") |
|
|
503 |
* cluster = '1', '-1' or 'NA' (for vertices not originally present in the |
|
|
504 |
original data) |
|
|
505 |
|
|
|
506 |
```{r} |
|
|
507 |
# graph cleaning |
|
|
508 |
graph_cleaning <- function(X, cluster.info){ |
|
|
509 |
# no reusability |
|
|
510 |
X <- igraph::simplify(X) |
|
|
511 |
va <- vertex_attr(X) |
|
|
512 |
viewed_mol <- c() |
|
|
513 |
for(omic in unique(cluster.info$block)){ |
|
|
514 |
mol <- intersect(cluster.info %>% dplyr::filter(.$block == omic) %>% |
|
|
515 |
pull(molecule), V(X)$name) |
|
|
516 |
viewed_mol <- c(viewed_mol, mol) |
|
|
517 |
X <- set_vertex_attr(graph = X, |
|
|
518 |
name = "type", |
|
|
519 |
index = mol, |
|
|
520 |
value = omic) |
|
|
521 |
X <- set_vertex_attr(graph = X, |
|
|
522 |
name = "mode", |
|
|
523 |
index = mol, |
|
|
524 |
value = "core") |
|
|
525 |
} |
|
|
526 |
# add medline ranker and go |
|
|
527 |
mol <- intersect(map(graph.go, ~ as_data_frame(.x)$to) %>% |
|
|
528 |
unlist %>% unique(), V(X)$name) # only GO terms |
|
|
529 |
viewed_mol <- c(viewed_mol, mol) |
|
|
530 |
X <- set_vertex_attr(graph = X, name = "type", index = mol, value = "GO") |
|
|
531 |
X <- set_vertex_attr(graph = X, name = "mode", |
|
|
532 |
index = mol, value = "extended") |
|
|
533 |
|
|
|
534 |
mol <- intersect(as.character(medlineranker.res.df$from), V(X)$name) |
|
|
535 |
viewed_mol <- c(viewed_mol, mol) |
|
|
536 |
X <- set_vertex_attr(graph = X, name = "type", |
|
|
537 |
index = mol, value = "Disease") |
|
|
538 |
X <- set_vertex_attr(graph = X, name = "mode", |
|
|
539 |
index = mol, value = "extended") |
|
|
540 |
|
|
|
541 |
other_mol <- setdiff(V(X), viewed_mol) |
|
|
542 |
if(!is_empty(other_mol)){ |
|
|
543 |
X <- set_vertex_attr(graph = X, name = "mode", |
|
|
544 |
index = other_mol, value = "extended") |
|
|
545 |
} |
|
|
546 |
X <- set_vertex_attr(graph = X, name = "mode", |
|
|
547 |
index = intersect(cluster.info$molecule, V(X)$name), |
|
|
548 |
value = "core") |
|
|
549 |
|
|
|
550 |
# signature |
|
|
551 |
mol <- intersect(V(X)$name, hmp_T2D$getCluster.sparse.res$molecule) |
|
|
552 |
X <- set_vertex_attr(graph = X, name = "sparse", index = mol, value = TRUE) |
|
|
553 |
mol <- setdiff(V(X)$name, hmp_T2D$getCluster.sparse.res$molecule) |
|
|
554 |
X <- set_vertex_attr(graph = X, name = "sparse", index = mol, value = FALSE) |
|
|
555 |
|
|
|
556 |
return(X) |
|
|
557 |
} |
|
|
558 |
``` |
|
|
559 |
|
|
|
560 |
|
|
|
561 |
```{r} |
|
|
562 |
FULL <- lapply(full.graph, function(x) graph_cleaning(x, cluster.info)) |
|
|
563 |
get_graph_stats(FULL) |
|
|
564 |
``` |
|
|
565 |
|
|
|
566 |
# Network exploration |
|
|
567 |
|
|
|
568 |
## Basics network exploration |
|
|
569 |
|
|
|
570 |
We can use basic graph statistics to explore the network such as degree |
|
|
571 |
distribution, modularity, and short path. |
|
|
572 |
|
|
|
573 |
```{r, eval = FALSE} |
|
|
574 |
# degree analysis |
|
|
575 |
d <- degree(FULL$All) |
|
|
576 |
hist(d) |
|
|
577 |
d[max(d)] |
|
|
578 |
|
|
|
579 |
# modularity # Warnings: can take several minutes |
|
|
580 |
res.mod <- walktrap.community(FULL$All) |
|
|
581 |
# ... |
|
|
582 |
|
|
|
583 |
# modularity |
|
|
584 |
sp <- shortest.paths(FULL$All) |
|
|
585 |
``` |
|
|
586 |
|
|
|
587 |
## Random walk with restart |
|
|
588 |
|
|
|
589 |
RWR is a powerful tool to explore the MO networks which simulates a particle |
|
|
590 |
that randomly walk on the network. |
|
|
591 |
From a starting point (`seed`) it ranks the other vertices based on their |
|
|
592 |
proximity with the seed and the network structure. |
|
|
593 |
|
|
|
594 |
We use RWR for function prediction and molecular mechanism identification. |
|
|
595 |
|
|
|
596 |
In the example below, the seeds were the GO terms vertices. |
|
|
597 |
|
|
|
598 |
```{r} |
|
|
599 |
# seeds = all vertices -> takes 5 minutes to run on regular computer |
|
|
600 |
# seeds <- V(FULL$All)$name |
|
|
601 |
# rwr_res <- random_walk_restart(FULL, seeds) |
|
|
602 |
|
|
|
603 |
# seed = some GO terms |
|
|
604 |
seeds <- head(V(FULL$All)$name[V(FULL$All)$type == "GO"]) |
|
|
605 |
rwr_res <- random_walk_restart(FULL, seeds) |
|
|
606 |
``` |
|
|
607 |
|
|
|
608 |
### Find vertices with specific attributes |
|
|
609 |
|
|
|
610 |
After the RWR analysis, we implemented several functions to extract valuable |
|
|
611 |
information. |
|
|
612 |
|
|
|
613 |
To identify MO molecular functions, the seed can be a GO term and we are |
|
|
614 |
interested to identify vertices with different omics type within the |
|
|
615 |
closest nodes. |
|
|
616 |
|
|
|
617 |
The function `rwr_find_seeds_between_attributes` can identify which seeds were |
|
|
618 |
able to reach vertices with different attributes (ex: `type`) within the |
|
|
619 |
closest `k` (ex: `15`) vertices. |
|
|
620 |
|
|
|
621 |
The function `summary_plot_rwr_attributes` displays the number of different |
|
|
622 |
values for a seed attribute as a bar graph. |
|
|
623 |
|
|
|
624 |
```{r} |
|
|
625 |
rwr_type_k15 <- rwr_find_seeds_between_attributes(X = rwr_res, |
|
|
626 |
attribute = "type", k = 15) |
|
|
627 |
|
|
|
628 |
# a summary plot function |
|
|
629 |
summary_plot_rwr_attributes(rwr_type_k15) |
|
|
630 |
summary_plot_rwr_attributes(rwr_type_k15$All) |
|
|
631 |
``` |
|
|
632 |
|
|
|
633 |
Alternatively, we can be interested to find functions or molecules which |
|
|
634 |
link different kinetic cluster (to find regulatory mechanisms). |
|
|
635 |
|
|
|
636 |
```{r} |
|
|
637 |
rwr_type_k15 <- rwr_find_seeds_between_attributes(X = rwr_res$All, |
|
|
638 |
attribute = "cluster", k = 15) |
|
|
639 |
summary_plot_rwr_attributes(rwr_type_k15) |
|
|
640 |
``` |
|
|
641 |
|
|
|
642 |
A RWR subnetworks can also be displayed with `plot_rwr_subnetwork` |
|
|
643 |
from a specific seed. |
|
|
644 |
```{r} |
|
|
645 |
sub_res <- rwr_type_k15$`GO:0005737` |
|
|
646 |
sub <- plot_rwr_subnetwork(sub_res, legend = TRUE, plot = TRUE) |
|
|
647 |
``` |
|
|
648 |
|
|
|
649 |
### Function prediction |
|
|
650 |
|
|
|
651 |
Finally, RWR can also be used for function prediction. |
|
|
652 |
From an annotated genes, the predicted function can be the closest vertex of the |
|
|
653 |
type "GO". |
|
|
654 |
|
|
|
655 |
We generalized this principle to identify, from a seed of interest, the closest |
|
|
656 |
node (or `top` closest nodes) with specific attributes and value. |
|
|
657 |
|
|
|
658 |
In the example below, the gene "ZNF263" is linked to the 5 closest nodes of |
|
|
659 |
type = 'GO' and type = 'Disease'. |
|
|
660 |
|
|
|
661 |
```{r} |
|
|
662 |
rwr_res <- random_walk_restart(FULL$All, seed = "ZNF263") |
|
|
663 |
|
|
|
664 |
# closest GO term |
|
|
665 |
rwr_find_closest_type(rwr_res, seed = "ZNF263", attribute = "type", |
|
|
666 |
value = "GO", top = 5) |
|
|
667 |
|
|
|
668 |
# closest Disease |
|
|
669 |
rwr_find_closest_type(rwr_res, seed = "ZNF263", attribute = "type", |
|
|
670 |
value = "Disease", top = 5) |
|
|
671 |
|
|
|
672 |
# closest nodes with an attribute "cluster" and the value "-1" |
|
|
673 |
rwr_find_closest_type(rwr_res, seed = "ZNF263", attribute = "cluster", |
|
|
674 |
value = "-1", top = 5) |
|
|
675 |
``` |
|
|
676 |
|
|
|
677 |
|
|
|
678 |
```{r, eval = FALSE} |
|
|
679 |
seeds <- V(FULL$All)$name[V(FULL$All)$type %in% c("GO", "Disease")] |
|
|
680 |
``` |
|
|
681 |
|
|
|
682 |
```{r} |
|
|
683 |
sessionInfo() |
|
|
684 |
``` |
|
|
685 |
|
|
|
686 |
# References |