-
Notifications
You must be signed in to change notification settings - Fork 14
/
multifactorial_analysis_BreastCancer.Rmd
1741 lines (1339 loc) · 78.7 KB
/
multifactorial_analysis_BreastCancer.Rmd
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
---
title: "MultiNicheNet analysis: anti-PD1 Breast cancer multifactorial comparison"
author: "Robin Browaeys"
package: "`r BiocStyle::pkg_ver('multinichenetr')`"
output:
BiocStyle::html_document
output_dir: "/Users/robinb/Work/multinichenetr/vignettes"
vignette: >
%\VignetteIndexEntry{MultiNicheNet analysis: anti-PD1 Breast cancer multifactorial comparison}
%\VignetteEngine{knitr::rmarkdown}
%\VignetteEncoding{UTF-8}
---
<style type="text/css">
.smaller {
font-size: 10px
}
</style>
<!-- github markdown built using
rmarkdown::render("vignettes/multifactorial_analysis_BreastCancer.Rmd", clean = FALSE )
-->
```{r setup, include = FALSE}
knitr::opts_chunk$set(
collapse = TRUE,
# comment = "#>",
warning = FALSE,
message = FALSE
)
library(BiocStyle)
```
In this vignette, you can learn how to perform a MultiNicheNet analysis on data with complex multifactorial experimental designs. More specifically, we will cover in depth how to study differential dynamics of cell-cell communication between conditions. We will assume the following type of design that is quite common: we have 2 or more conditions/groups of interest, and for each condition we also have 2 or more time points (eg before and after a certain treatment). Research questions are: how does cell-cell communication change over time in each condition? And, importantly, how are these time-related changes different between the conditions? Certainly the latter is a non-trivial question. We will here demonstrate how MultiNicheNet can exploit the flexibility of generalized linear models in the pseudobulk-edgeR framework to handle this non-trivial question.
This vignette is quite advanced, so if you are new to MultiNicheNet, we recommend reading and running this vignette: [basis_analysis_steps_MISC.knit.md](basis_analysis_steps_MISC.knit.md) to get acquainted with the methodology and simple applications.
As example expression data of interacting cells for this vignette, we will here use scRNAseq data from breast cancer biopsies of patients receiving anti-PD1 immune-checkpoint blockade therapy. Bassez et al. collected from each patient one tumor biopsy before anti-PD1 therapy (“pre-treatment”) and one during subsequent surgery (“on-treatment”) [A single-cell map of intratumoral changes during anti-PD1 treatment of patients with breast cancer](https://www.nature.com/articles/s41591-021-01323-8). Based on additional scTCR-seq results, they identified one group of patients with clonotype expansion as response to the therapy (“E”) and one group with only limited or no clonotype expansion (“NE”). To make this vignette more easily adaptable to the user, we will rename the groups and time points as follows: E --> group1, NE --> group2, Pre-treatment: timepoint1, On-treatment: timepoint2.
Now, how can we define which CCC patterns are changing during anti-PD1 therapy and which of these changes are specific to E compared to NE patients (and vice versa)?
First, we will run a MultiNicheNet, focusing on anti-PD1 therapy related differences in group1. Secondly, we will do the same for group2. Then we will compare the prioritized interactions between both groups in a basic way. Finally, we will run a MultiNicheNet analysis with a complex contrast to focus specifically on group-specific differences in anti-PD1 therapy associated CCC changes.
In this vignette, we will first prepare the common elements of all MultiNicheNet core analyses, then run, interpret and compare the different analyses.
# Preparation of the MultiNicheNet core analysis
```{r load-libs, message = FALSE, warning = FALSE}
library(SingleCellExperiment)
library(dplyr)
library(ggplot2)
library(nichenetr)
library(multinichenetr)
```
## Load NicheNet's ligand-receptor network and ligand-target matrix
The Nichenet v2 networks and matrices for both mouse and human can be downloaded from Zenodo [![DOI](https://zenodo.org/badge/DOI/10.5281/zenodo.7074291.svg)](https://doi.org/10.5281/zenodo.7074291).
We will read these object in for human because our expression data is of human patients.
Gene names are here made syntactically valid via `make.names()` to avoid the loss of genes (eg H2-M3) in downstream visualizations.
```{r}
organism = "human"
```
```{r, results='hide'}
options(timeout = 250)
if(organism == "human"){
lr_network_all =
readRDS(url(
"https://zenodo.org/record/10229222/files/lr_network_human_allInfo_30112033.rds"
)) %>%
mutate(
ligand = convert_alias_to_symbols(ligand, organism = organism),
receptor = convert_alias_to_symbols(receptor, organism = organism))
lr_network_all = lr_network_all %>%
mutate(ligand = make.names(ligand), receptor = make.names(receptor))
lr_network = lr_network_all %>%
distinct(ligand, receptor)
ligand_target_matrix = readRDS(url(
"https://zenodo.org/record/7074291/files/ligand_target_matrix_nsga2r_final.rds"
))
colnames(ligand_target_matrix) = colnames(ligand_target_matrix) %>%
convert_alias_to_symbols(organism = organism) %>% make.names()
rownames(ligand_target_matrix) = rownames(ligand_target_matrix) %>%
convert_alias_to_symbols(organism = organism) %>% make.names()
lr_network = lr_network %>% filter(ligand %in% colnames(ligand_target_matrix))
ligand_target_matrix = ligand_target_matrix[, lr_network$ligand %>% unique()]
} else if(organism == "mouse"){
lr_network_all = readRDS(url(
"https://zenodo.org/record/10229222/files/lr_network_mouse_allInfo_30112033.rds"
)) %>%
mutate(
ligand = convert_alias_to_symbols(ligand, organism = organism),
receptor = convert_alias_to_symbols(receptor, organism = organism))
lr_network_all = lr_network_all %>%
mutate(ligand = make.names(ligand), receptor = make.names(receptor))
lr_network = lr_network_all %>%
distinct(ligand, receptor)
ligand_target_matrix = readRDS(url(
"https://zenodo.org/record/7074291/files/ligand_target_matrix_nsga2r_final_mouse.rds"
))
colnames(ligand_target_matrix) = colnames(ligand_target_matrix) %>%
convert_alias_to_symbols(organism = organism) %>% make.names()
rownames(ligand_target_matrix) = rownames(ligand_target_matrix) %>%
convert_alias_to_symbols(organism = organism) %>% make.names()
lr_network = lr_network %>% filter(ligand %in% colnames(ligand_target_matrix))
ligand_target_matrix = ligand_target_matrix[, lr_network$ligand %>% unique()]
}
```
## Read in SingleCellExperiment Object
In this vignette, we will load in a subset of the breast cancer scRNAseq data [![DOI](https://zenodo.org/badge/DOI/10.5281/zenodo.8010790.svg)](https://doi.org/10.5281/zenodo.8010790). For the sake of demonstration, this subset only contains 3 cell types.
Because the NicheNet 2.0. networks are in the most recent version of the official gene symbols, we will make sure that the gene symbols used in the expression data are also updated (= converted from their "aliases" to official gene symbols). Afterwards, we will make them again syntactically valid.
```{r}
sce = readRDS(url(
"https://zenodo.org/record/8010790/files/sce_subset_breastcancer.rds"
))
sce = alias_to_symbol_SCE(sce, "human") %>% makenames_SCE()
```
Now we will go further in defining the settings for the MultiNicheNet analysis
## Prepare the settings of the MultiNicheNet cell-cell communication analysis
In this step, we will formalize our research question into MultiNicheNet input arguments.
In this case study, we want to study differences in therapy-induced cell-cell communication changes (On-vs-Pre therapy) between two patient groups (E vs NE: patients with clonotype expansion versus patients without clonotype expansion). Both therapy-timepoint and patient group are indicated in the following meta data column: `expansion_timepoint`, which has 4 different values: PreE, PreNE, OnE, OnNE.
Cell type annotations are indicated in the `subType` column, and the sample is indicated by the `sample_id` column.
```{r}
sample_id = "sample_id"
group_id = "expansion_timepoint"
celltype_id = "subType"
```
__Important__: It is required that each sample-id is uniquely assigned to only one condition/group of interest. Therefore, our `sample_id` here does not only indicate the patient, but also the timepoint of sampling.
If you would have batch effects or covariates you can correct for, you can define this here as well. However, this is not applicable to this dataset. Therefore we will use the following NA settings:
```{r}
batches = NA
covariates = NA
```
__Important__: for batches, there should be at least one sample for every group-batch combination. If one of your groups/conditions lacks a certain level of your batch, you won't be able to correct for the batch effect because the model is then not able to distinguish batch from group/condition effects.
__Important__: The column names of group, sample, cell type, and batches should be syntactically valid (`make.names`)
__Important__: All group, sample, cell type, and batch names should be syntactically valid as well (`make.names`) (eg through `SummarizedExperiment::colData(sce)$ShortID = SummarizedExperiment::colData(sce)$ShortID %>% make.names()`)
If you want to focus the analysis on specific cell types (e.g. because you know which cell types reside in the same microenvironments based on spatial data), you can define this here. If you have sufficient computational resources and no specific idea of cell-type colocalzations, we recommend to consider all cell types as potential senders and receivers.
Here we will consider all cell types in the data:
```{r}
senders_oi = SummarizedExperiment::colData(sce)[,celltype_id] %>% unique()
receivers_oi = SummarizedExperiment::colData(sce)[,celltype_id] %>% unique()
sce = sce[, SummarizedExperiment::colData(sce)[,celltype_id] %in%
c(senders_oi, receivers_oi)
]
```
In case you would have samples in your data that do not belong to one of the groups/conditions of interest, we recommend removing them and only keeping conditions of interest.
Before we do this, we will here rename our group/timepoint combinations.
This is not necessary for you to do this, but then you will need to replace "G1.T1" with the equivalent of group1-timepoint1 in your data. Etc.
```{r}
SummarizedExperiment::colData(sce)[,group_id][SummarizedExperiment::colData(sce)[,group_id] == "PreE"] = "G1.T1"
SummarizedExperiment::colData(sce)[,group_id][SummarizedExperiment::colData(sce)[,group_id] == "PreNE"] = "G2.T1"
SummarizedExperiment::colData(sce)[,group_id][SummarizedExperiment::colData(sce)[,group_id] == "OnE"] = "G1.T2"
SummarizedExperiment::colData(sce)[,group_id][SummarizedExperiment::colData(sce)[,group_id] == "OnNE"] = "G2.T2"
```
```{r}
conditions_keep = c("G1.T1", "G2.T1", "G1.T2", "G2.T2")
sce = sce[, SummarizedExperiment::colData(sce)[,group_id] %in%
conditions_keep
]
```
## Defining the parameters of the MultiNicheNet core analyses
In MultiNicheNet, following steps are executed:
* 1. Cell-type filtering: determine which cell types are sufficiently present
* 2. Gene filtering: determine which genes are sufficiently expressed in each present cell type
* 3. Pseudobulk expression calculation: determine and normalize per-sample pseudobulk expression levels for each expressed gene in each present cell type
* 4. Differential expression (DE) analysis: determine which genes are differentially expressed
* 5. Ligand activity prediction: use the DE analysis output to predict the activity of ligands in receiver cell types and infer their potential target genes
* 6. Prioritization: rank cell-cell communication patterns through multi-criteria prioritization
For each of these steps, some parameters need to be defined. This is what we will do now.
*Parameters for step 1: Cell-type filtering:*
Since MultiNicheNet will infer group differences at the sample level for each cell type (currently via Muscat - pseudobulking + EdgeR), we need to have sufficient cells per sample of a cell type, and this for all groups. In the following analysis we will set this minimum number of cells per cell type per sample at 10. Samples that have less than `min_cells` cells will be excluded from the analysis for that specific cell type.
```{r}
min_cells = 10
```
*Parameters for step 2: Gene filtering*
For each cell type, we will consider genes expressed if they are expressed in at least a `min_sample_prop` fraction of samples in the condition with the lowest number of samples. By default, we set `min_sample_prop = 0.50`.
```{r}
min_sample_prop = 0.50
```
But how do we define which genes are expressed in a sample? For this we will consider genes as expressed if they have non-zero expression values in a `fraction_cutoff` fraction of cells of that cell type in that sample. By default, we set `fraction_cutoff = 0.05`, which means that genes should show non-zero expression values in at least 5% of cells in a sample.
```{r}
fraction_cutoff = 0.05
```
*Parameters for step 5: Ligand activity prediction*
One of the prioritization criteria is the predicted activity of ligands in receiver cell types. Similarly to base NicheNet (https://github.com/saeyslab/nichenetr), we use the DE output to create a "geneset of interest": here we assume that DE genes within a cell type may be DE because of differential cell-cell communication processes. To determine the genesets of interest based on DE output, we need to define which logFC and/or p-value thresholds we will use.
By default, we will apply the p-value cutoff on the normal p-values, and not on the p-values corrected for multiple testing. This choice was made because most multi-sample single-cell transcriptomics datasets have just a few samples per group and we might have a lack of statistical power due to pseudobulking. But, if the smallest group >= 20 samples, we typically recommend using p_val_adj = TRUE. When the biological difference between the conditions is very large, we typically recommend increasing the logFC_threshold and/or using p_val_adj = TRUE.
```{r}
logFC_threshold = 0.50
p_val_threshold = 0.05
```
```{r}
p_val_adj = FALSE
```
After the ligand activity prediction, we will also infer the predicted target genes of these ligands in each contrast. For this ligand-target inference procedure, we also need to select which top n of the predicted target genes will be considered (here: top 250 targets per ligand). This parameter will not affect the ligand activity predictions. It will only affect ligand-target visualizations and construction of the intercellular regulatory network during the downstream analysis. We recommend users to test other settings in case they would be interested in exploring fewer, but more confident target genes, or vice versa.
```{r}
top_n_target = 250
```
The NicheNet ligand activity analysis can be run in parallel for each receiver cell type, by changing the number of cores as defined here. Using more cores will speed up the analysis at the cost of needing more memory. This is only recommended if you have many receiver cell types of interest. You can define here the maximum number of cores you want to be used.
```{r}
n.cores = 8
```
*Parameters for step 6: Prioritization*
We will use the following criteria to prioritize ligand-receptor interactions:
* Upregulation of the ligand in a sender cell type and/or upregulation of the receptor in a receiver cell type - in the condition of interest.
* Cell-type specific expression of the ligand in the sender cell type and receptor in the receiver cell type in the condition of interest (to mitigate the influence of upregulated but still relatively weakly expressed ligands/receptors).
* Sufficiently high expression levels of ligand and receptor in many samples of the same group.
* High NicheNet ligand activity, to further prioritize ligand-receptor pairs based on their predicted effect of the ligand-receptor interaction on the gene expression in the receiver cell type.
We will combine these prioritization criteria in a single aggregated prioritization score. In the default setting, we will weigh each of these criteria equally (`scenario = "regular"`). This setting is strongly recommended. However, we also provide some additional setting to accomodate different biological scenarios. The setting `scenario = "lower_DE"` halves the weight for DE criteria and doubles the weight for ligand activity. This is recommended in case your hypothesis is that the differential CCC patterns in your data are less likely to be driven by DE (eg in cases of differential migration into a niche). The setting `scenario = "no_frac_LR_expr"` ignores the criterion "Sufficiently high expression levels of ligand and receptor in many samples of the same group". This may be interesting for users that have data with a limited number of samples and don’t want to penalize interactions if they are not sufficiently expressed in some samples.
Here we will choose for the regular setting.
```{r}
scenario = "regular"
```
Finally, we still need to make one choice. For NicheNet ligand activity we can choose to prioritize ligands that only induce upregulation of target genes (`ligand_activity_down = FALSE`) or can lead potentially lead to both up- and downregulation (`ligand_activity_down = TRUE`). The benefit of `ligand_activity_down = FALSE` is ease of interpretability: prioritized ligand-receptor pairs will be upregulated in the condition of interest, just like their target genes. `ligand_activity_down = TRUE` can be harder to interpret because target genes of some interactions may be upregulated in the other conditions compared to the condition of interest. This is harder to interpret, but may help to pick up interactions that can also repress gene expression.
Here we will choose for setting `ligand_activity_down = FALSE` and focus specifically on upregulating ligands.
```{r}
ligand_activity_down = FALSE
```
# Running the MultiNicheNet core analyses
## 1) G1.T2-G1.T1: Anti-PD1 therapy associated cell-cell communication changes in patients with clonotype expansion (E)
*Set the required contrast of interest.*
Here, we want to compare on-treatment samples with pre-treatment samples for group E. Therefore we can set this contrast as:
```{r}
contrasts_oi = c("'G1.T2-G1.T1','G1.T1-G1.T2'")
```
__Very Important__ Note the format to indicate the contrasts! This formatting should be adhered to very strictly, and white spaces are not allowed! Check `?get_DE_info` for explanation about how to define this well. The most important points are that:
*each contrast is surrounded by single quotation marks
*contrasts are separated by a comma without any white space
*all contrasts together are surrounded by double quotation marks.
For downstream visualizations and linking contrasts to their main condition, we also need to run the following:
This is necessary because we will also calculate cell-type+condition specificity of ligands and receptors.
```{r}
contrast_tbl = tibble(
contrast = c("G1.T2-G1.T1", "G1.T1-G1.T2"),
group = c("G1.T2","G1.T1")
)
```
### Run the analysis
*Cell-type filtering: determine which cell types are sufficiently present*
```{r}
abundance_info = get_abundance_info(
sce = sce,
sample_id = sample_id, group_id = group_id, celltype_id = celltype_id,
min_cells = min_cells,
senders_oi = senders_oi, receivers_oi = receivers_oi,
batches = batches
)
```
```{r, fig.width=10}
abundance_info$abund_plot_sample
```
*Gene filtering: determine which genes are sufficiently expressed in each present cell type*
```{r}
frq_list = get_frac_exprs(
sce = sce,
sample_id = sample_id, celltype_id = celltype_id, group_id = group_id,
batches = batches,
min_cells = min_cells,
fraction_cutoff = fraction_cutoff, min_sample_prop = min_sample_prop)
```
Now only keep genes that are expressed by at least one cell type:
```{r}
genes_oi = frq_list$expressed_df %>%
filter(expressed == TRUE) %>% pull(gene) %>% unique()
sce = sce[genes_oi, ]
```
*Pseudobulk expression calculation: determine and normalize per-sample pseudobulk expression levels for each expressed gene in each present cell type*
```{r}
abundance_expression_info = process_abundance_expression_info(
sce = sce,
sample_id = sample_id, group_id = group_id, celltype_id = celltype_id,
min_cells = min_cells,
senders_oi = senders_oi, receivers_oi = receivers_oi,
lr_network = lr_network,
batches = batches,
frq_list = frq_list,
abundance_info = abundance_info)
```
*Differential expression (DE) analysis: determine which genes are differentially expressed*
```{r}
DE_info_group1 = get_DE_info(
sce = sce,
sample_id = sample_id, group_id = group_id, celltype_id = celltype_id,
batches = batches, covariates = covariates,
contrasts_oi = contrasts_oi,
min_cells = min_cells,
expressed_df = frq_list$expressed_df)
```
Check DE output information in table with logFC and p-values for each gene-celltype-contrast:
```{r}
DE_info_group1$celltype_de$de_output_tidy %>% head()
```
Evaluate the distributions of p-values:
```{r}
DE_info_group1$hist_pvals
```
These distributions look fine (uniform distribution, except peak at p-value <= 0.05), so we will continue using these regular p-values. In case these p-value distributions look irregular, you can estimate empirical p-values as we will demonstrate in another vignette.
```{r}
empirical_pval = FALSE
```
```{r}
if(empirical_pval == TRUE){
DE_info_emp = get_empirical_pvals(DE_info_group1$celltype_de$de_output_tidy)
celltype_de_group1 = DE_info_emp$de_output_tidy_emp %>% select(-p_val, -p_adj) %>%
rename(p_val = p_emp, p_adj = p_adj_emp)
} else {
celltype_de_group1 = DE_info_group1$celltype_de$de_output_tidy
}
```
*Combine DE information for ligand-senders and receptors-receivers*
```{r}
sender_receiver_de = combine_sender_receiver_de(
sender_de = celltype_de_group1,
receiver_de = celltype_de_group1,
senders_oi = senders_oi,
receivers_oi = receivers_oi,
lr_network = lr_network
)
```
```{r}
sender_receiver_de %>% head(20)
```
*Ligand activity prediction: use the DE analysis output to predict the activity of ligands in receiver cell types and infer their potential target genes*
We will first inspect the geneset_oi-vs-background ratios for the default tresholds:
```{r}
geneset_assessment = contrast_tbl$contrast %>%
lapply(
process_geneset_data,
celltype_de_group1, logFC_threshold, p_val_adj, p_val_threshold
) %>%
bind_rows()
geneset_assessment
```
We can see here that for all cell type / contrast combinations, all geneset/background ratio's are within the recommended range (`in_range_up` and `in_range_down` columns). When these geneset/background ratio's would not be within the recommended ranges, we should interpret ligand activity results for these cell types with more caution, or use different thresholds (for these or all cell types).
```{r}
ligand_activities_targets_DEgenes = suppressMessages(suppressWarnings(
get_ligand_activities_targets_DEgenes(
receiver_de = celltype_de_group1,
receivers_oi = intersect(receivers_oi, celltype_de_group1$cluster_id %>% unique()),
ligand_target_matrix = ligand_target_matrix,
logFC_threshold = logFC_threshold,
p_val_threshold = p_val_threshold,
p_val_adj = p_val_adj,
top_n_target = top_n_target,
verbose = TRUE,
n.cores = n.cores
)
))
```
*Prioritization: rank cell-cell communication patterns through multi-criteria prioritization*
```{r}
sender_receiver_tbl = sender_receiver_de %>% distinct(sender, receiver)
metadata_combined = SummarizedExperiment::colData(sce) %>% tibble::as_tibble()
if(!is.na(batches)){
grouping_tbl = metadata_combined[,c(sample_id, group_id, batches)] %>%
tibble::as_tibble() %>% distinct()
colnames(grouping_tbl) = c("sample","group",batches)
} else {
grouping_tbl = metadata_combined[,c(sample_id, group_id)] %>%
tibble::as_tibble() %>% distinct()
colnames(grouping_tbl) = c("sample","group")
}
prioritization_tables = suppressMessages(generate_prioritization_tables(
sender_receiver_info = abundance_expression_info$sender_receiver_info,
sender_receiver_de = sender_receiver_de,
ligand_activities_targets_DEgenes = ligand_activities_targets_DEgenes,
contrast_tbl = contrast_tbl,
sender_receiver_tbl = sender_receiver_tbl,
grouping_tbl = grouping_tbl,
scenario = "regular", # all prioritization criteria will be weighted equally
fraction_cutoff = fraction_cutoff,
abundance_data_receiver = abundance_expression_info$abundance_data_receiver,
abundance_data_sender = abundance_expression_info$abundance_data_sender,
ligand_activity_down = ligand_activity_down
))
```
*Compile the MultiNicheNet output object*
```{r}
multinichenet_output_group1_t2vst1 = list(
celltype_info = abundance_expression_info$celltype_info,
celltype_de = celltype_de_group1,
sender_receiver_info = abundance_expression_info$sender_receiver_info,
sender_receiver_de = sender_receiver_de,
ligand_activities_targets_DEgenes = ligand_activities_targets_DEgenes,
prioritization_tables = prioritization_tables,
grouping_tbl = grouping_tbl,
lr_target_prior_cor = tibble()
)
multinichenet_output_group1_t2vst1 = make_lite_output(multinichenet_output_group1_t2vst1)
```
### Interpret the analysis
*Summarizing ChordDiagram circos plots*
In a first instance, we will look at the broad overview of prioritized interactions via condition-specific Chordiagram circos plots.
We will look here at the top 50 predictions across all contrasts, senders, and receivers of interest.
```{r}
prioritized_tbl_oi_all = get_top_n_lr_pairs(
multinichenet_output_group1_t2vst1$prioritization_tables,
top_n = 50,
rank_per_group = FALSE
)
```
```{r, fig.width=12, fig.height=12}
prioritized_tbl_oi =
multinichenet_output_group1_t2vst1$prioritization_tables$group_prioritization_tbl %>%
filter(id %in% prioritized_tbl_oi_all$id) %>%
distinct(id, sender, receiver, ligand, receptor, group) %>%
left_join(prioritized_tbl_oi_all)
prioritized_tbl_oi$prioritization_score[is.na(prioritized_tbl_oi$prioritization_score)] = 0
senders_receivers = union(prioritized_tbl_oi$sender %>% unique(), prioritized_tbl_oi$receiver %>% unique()) %>% sort()
colors_sender = RColorBrewer::brewer.pal(n = length(senders_receivers), name = 'Spectral') %>% magrittr::set_names(senders_receivers)
colors_receiver = RColorBrewer::brewer.pal(n = length(senders_receivers), name = 'Spectral') %>% magrittr::set_names(senders_receivers)
circos_list = make_circos_group_comparison(prioritized_tbl_oi, colors_sender, colors_receiver)
```
We observe more interactions that increase during therapy ("G1.T2" condition) compared to interactions that decrease ("G1.T1" condition)
*Interpretable bubble plots*
Whereas these ChordDiagrams show the most specific interactions per group, they don't give insights into the data behind these predictions. Therefore we will now look at visualizations that indicate the different prioritization criteria used in MultiNicheNet.
In the next type of plots, we will 1) visualize the per-sample scaled product of normalized ligand and receptor pseudobulk expression, 2) visualize the scaled ligand activities, and 3) cell-type specificity.
We will now check the top interactions specific for the OnE group (G1.T2) - so these are the interactions that increase during anti-PD1 therapy.
```{r}
group_oi = "G1.T2"
```
```{r}
prioritized_tbl_oi_50 = get_top_n_lr_pairs(
multinichenet_output_group1_t2vst1$prioritization_tables,
top_n = 50, rank_per_group = FALSE) %>% filter(group == group_oi)
```
```{r}
prioritized_tbl_oi_50_omnipath = prioritized_tbl_oi_50 %>%
inner_join(lr_network_all)
```
Now we add this to the bubble plot visualization:
```{r, fig.height=13, fig.width=24}
plot_oi = make_sample_lr_prod_activity_plots_Omnipath(
multinichenet_output_group1_t2vst1$prioritization_tables,
prioritized_tbl_oi_50_omnipath)
plot_oi
```
Interestingly, we can already see that some of these interactions are only increasing in the E group, whereas others change in the NE group as well!
In practice, we always recommend generating these plots for all conditions/contrasts. To avoid that this vignette becomes too long, we will not do this here.
Now let's do the same analysis for the NE group instead
## 2) G2.T2-G2.T1: Anti-PD1 therapy associated cell-cell communication changes in patients without clonotype expansion (NE)
*Set the required contrast of interest.*
Here, we want to compare on-treatment samples with pre-treatment samples for group NE. Therefore we can set this contrast as:
```{r}
contrasts_oi = c("'G2.T2-G2.T1','G2.T1-G2.T2'")
```
```{r}
contrast_tbl = tibble(
contrast = c("G2.T2-G2.T1", "G2.T1-G2.T2"),
group = c("G2.T2","G2.T1")
)
```
### Run the analysis
In this case, the first 3 steps (cell-type filtering, gene filtering, and pseudobulk expression calculation) were run for the entire dataset and are exactly the same now as for the first MultiNicheNet analysis. Therefore, we will not redo these steps, and just start with step 4, the first unique step for this second analysis.
*Differential expression (DE) analysis: determine which genes are differentially expressed*
```{r}
DE_info_group2 = get_DE_info(
sce = sce,
sample_id = sample_id, group_id = group_id, celltype_id = celltype_id,
batches = batches, covariates = covariates,
contrasts_oi = contrasts_oi,
min_cells = min_cells,
expressed_df = frq_list$expressed_df)
```
Check DE output information in table with logFC and p-values for each gene-celltype-contrast:
```{r}
DE_info_group2$celltype_de$de_output_tidy %>% head()
```
Evaluate the distributions of p-values:
```{r}
DE_info_group2$hist_pvals
```
These distributions look fine (uniform distribution, except peak at p-value <= 0.05), so we will continue using these regular p-values. In case these p-value distributions look irregular, you can estimate empirical p-values as we will demonstrate in another vignette.
```{r}
empirical_pval = FALSE
```
```{r}
if(empirical_pval == TRUE){
DE_info_emp = get_empirical_pvals(DE_info_group2$celltype_de$de_output_tidy)
celltype_de_group2 = DE_info_emp$de_output_tidy_emp %>% select(-p_val, -p_adj) %>%
rename(p_val = p_emp, p_adj = p_adj_emp)
} else {
celltype_de_group2 = DE_info_group2$celltype_de$de_output_tidy
}
```
*Combine DE information for ligand-senders and receptors-receivers*
```{r}
sender_receiver_de = combine_sender_receiver_de(
sender_de = celltype_de_group2,
receiver_de = celltype_de_group2,
senders_oi = senders_oi,
receivers_oi = receivers_oi,
lr_network = lr_network
)
```
```{r}
sender_receiver_de %>% head(20)
```
*Ligand activity prediction: use the DE analysis output to predict the activity of ligands in receiver cell types and infer their potential target genes*
We will first inspect the geneset_oi-vs-background ratios for the default tresholds:
```{r}
geneset_assessment = contrast_tbl$contrast %>%
lapply(
process_geneset_data,
celltype_de_group2, logFC_threshold, p_val_adj, p_val_threshold
) %>%
bind_rows()
geneset_assessment
```
We can see here that for all cell type / contrast combinations, all geneset/background ratio's are within the recommended range (`in_range_up` and `in_range_down` columns). When these geneset/background ratio's would not be within the recommended ranges, we should interpret ligand activity results for these cell types with more caution, or use different thresholds (for these or all cell types).
```{r}
ligand_activities_targets_DEgenes = suppressMessages(suppressWarnings(
get_ligand_activities_targets_DEgenes(
receiver_de = celltype_de_group2,
receivers_oi = intersect(receivers_oi, celltype_de_group2$cluster_id %>% unique()),
ligand_target_matrix = ligand_target_matrix,
logFC_threshold = logFC_threshold,
p_val_threshold = p_val_threshold,
p_val_adj = p_val_adj,
top_n_target = top_n_target,
verbose = TRUE,
n.cores = n.cores
)
))
```
*Prioritization: rank cell-cell communication patterns through multi-criteria prioritization*
```{r}
sender_receiver_tbl = sender_receiver_de %>% distinct(sender, receiver)
metadata_combined = SummarizedExperiment::colData(sce) %>% tibble::as_tibble()
if(!is.na(batches)){
grouping_tbl = metadata_combined[,c(sample_id, group_id, batches)] %>%
tibble::as_tibble() %>% distinct()
colnames(grouping_tbl) = c("sample","group",batches)
} else {
grouping_tbl = metadata_combined[,c(sample_id, group_id)] %>%
tibble::as_tibble() %>% distinct()
colnames(grouping_tbl) = c("sample","group")
}
prioritization_tables = suppressMessages(generate_prioritization_tables(
sender_receiver_info = abundance_expression_info$sender_receiver_info,
sender_receiver_de = sender_receiver_de,
ligand_activities_targets_DEgenes = ligand_activities_targets_DEgenes,
contrast_tbl = contrast_tbl,
sender_receiver_tbl = sender_receiver_tbl,
grouping_tbl = grouping_tbl,
scenario = "regular", # all prioritization criteria will be weighted equally
fraction_cutoff = fraction_cutoff,
abundance_data_receiver = abundance_expression_info$abundance_data_receiver,
abundance_data_sender = abundance_expression_info$abundance_data_sender,
ligand_activity_down = ligand_activity_down
))
```
*Compile the MultiNicheNet output object*
```{r}
multinichenet_output_group2_t2vst1 = list(
celltype_info = abundance_expression_info$celltype_info,
celltype_de = celltype_de_group2,
sender_receiver_info = abundance_expression_info$sender_receiver_info,
sender_receiver_de = sender_receiver_de,
ligand_activities_targets_DEgenes = ligand_activities_targets_DEgenes,
prioritization_tables = prioritization_tables,
grouping_tbl = grouping_tbl,
lr_target_prior_cor = tibble()
)
multinichenet_output_group2_t2vst1 = make_lite_output(multinichenet_output_group2_t2vst1)
```
### Interpret the analysis
*Summarizing ChordDiagram circos plots*
We will look here at the top 50 predictions across all contrasts, senders, and receivers of interest.
```{r}
prioritized_tbl_oi_all = get_top_n_lr_pairs(
multinichenet_output_group2_t2vst1$prioritization_tables,
top_n = 50,
rank_per_group = FALSE
)
```
```{r, fig.width=12, fig.height=12}
prioritized_tbl_oi =
multinichenet_output_group2_t2vst1$prioritization_tables$group_prioritization_tbl %>%
filter(id %in% prioritized_tbl_oi_all$id) %>%
distinct(id, sender, receiver, ligand, receptor, group) %>%
left_join(prioritized_tbl_oi_all)
prioritized_tbl_oi$prioritization_score[is.na(prioritized_tbl_oi$prioritization_score)] = 0
senders_receivers = union(prioritized_tbl_oi$sender %>% unique(), prioritized_tbl_oi$receiver %>% unique()) %>% sort()
colors_sender = RColorBrewer::brewer.pal(n = length(senders_receivers), name = 'Spectral') %>% magrittr::set_names(senders_receivers)
colors_receiver = RColorBrewer::brewer.pal(n = length(senders_receivers), name = 'Spectral') %>% magrittr::set_names(senders_receivers)
circos_list = make_circos_group_comparison(prioritized_tbl_oi, colors_sender, colors_receiver)
```
Also in this group of patients we see rather an increase on therapy compared to pre-therapy.
*Interpretable bubble plots*
We will now check the top interactions specific for the OnE group - so these are the interactions that increase during anti-PD1 therapy.
```{r}
group_oi = "G2.T2"
```
```{r}
prioritized_tbl_oi_50 = get_top_n_lr_pairs(
multinichenet_output_group2_t2vst1$prioritization_tables,
top_n = 50, rank_per_group = FALSE) %>% filter(group == group_oi)
```
```{r}
prioritized_tbl_oi_50_omnipath = prioritized_tbl_oi_50 %>%
inner_join(lr_network_all)
```
Now we add this to the bubble plot visualization:
```{r, fig.height=11, fig.width=24}
plot_oi = make_sample_lr_prod_activity_plots_Omnipath(
multinichenet_output_group2_t2vst1$prioritization_tables,
prioritized_tbl_oi_50_omnipath)
plot_oi
```
So interesting observation: even though the NE group is characterized by a lack of T cell clonotype expansion after therapy, there are still clear therapy-induced differences on gene expression and cell-cell communication.
Interestingly, we can see here as well that some interactions are only increasing in the NE group, whereas others change in the E group as well!
This brings us to the next research question: do some interactions increase/decrease more in the E group vs the NE group or vice versa?
## 3A) E-vs-NE: simple comparison of prioritized anti-PD1 therapy-associated CCC patterns
To address this question, we will first do a simple analysis. We will just compare the prioritization scores between the E and the NE group, to find CCC patterns that are most different between the E and NE group.
*Set up the comparisons*
```{r}
comparison_table = multinichenet_output_group1_t2vst1$prioritization_tables$group_prioritization_tbl %>%
select(group, id, prioritization_score) %>%
rename(prioritization_score_group1 = prioritization_score) %>%
mutate(group = case_when(
group == "G1.T1" ~ "timepoint1",
group == "G1.T2" ~ "timepoint2",
TRUE ~ group)
) %>%
inner_join(
multinichenet_output_group2_t2vst1$prioritization_tables$group_prioritization_tbl %>%
select(group, id, prioritization_score) %>%
rename(prioritization_score_group2 = prioritization_score) %>%
mutate(group = case_when(
group == "G2.T1" ~ "timepoint1",
group == "G2.T2" ~ "timepoint2",
TRUE ~ group)
)
) %>%
mutate(difference_group1_vs_group2 = prioritization_score_group1 - prioritization_score_group2)
```
Let's now inspect some of the top interactions that got much higher scores in the E group
```{r}
comparison_table %>%
arrange(-difference_group1_vs_group2) %>%
filter(prioritization_score_group1 > 0.80)
```
Let's now inspect some of the top interactions that got much higher scores in the NE group
```{r}
comparison_table %>%
arrange(difference_group1_vs_group2) %>%
filter(prioritization_score_group2 > 0.80)
```
### Interactions increasing after therapy in the E group
Visualize now interactions that are in the top250 interactions for the contrast On-vs-Pre in the E group.
*E-group specific*
Of these, we will first zoom into the top25 interactions with biggest difference in the E-vs-NE group.
```{r}
group_oi = "G1.T2"
```
```{r}
prioritized_tbl_oi_250 = get_top_n_lr_pairs(
multinichenet_output_group1_t2vst1$prioritization_tables,
top_n = 250, rank_per_group = FALSE) %>%
filter(group == group_oi) %>%
inner_join(lr_network_all)
```
```{r}
ids_group1_vs_group2 = comparison_table %>% filter(group == "timepoint2" & id %in% prioritized_tbl_oi_250$id) %>% top_n(25, difference_group1_vs_group2) %>% filter(difference_group1_vs_group2 > 0) %>% pull(id)
```
```{r}
prioritized_tbl_oi_250_unique_E = prioritized_tbl_oi_250 %>%
filter(id %in% ids_group1_vs_group2)
```
```{r, fig.height=7, fig.width=24}
plot_oi = make_sample_lr_prod_activity_plots_Omnipath(
multinichenet_output_group1_t2vst1$prioritization_tables,
prioritized_tbl_oi_250_unique_E)
plot_oi
```
*shared between E-group and NE-group*
Now we will look at the interactions with a very similar score between E and NE. So these interactions are interactions that increase after therapy, but shared between group E and NE.
```{r}
ids_E_shared_NE = comparison_table %>% filter(group == "timepoint2" & id %in% prioritized_tbl_oi_250$id) %>% mutate(deviation_from_0 = abs(0-difference_group1_vs_group2)) %>% top_n(25, -deviation_from_0) %>% arrange(deviation_from_0) %>% pull(id)
```
```{r}
prioritized_tbl_oi_250_shared_E_NE = prioritized_tbl_oi_250 %>%
filter(id %in% ids_E_shared_NE)
```
```{r, fig.height=7, fig.width=24}
plot_oi = make_sample_lr_prod_activity_plots_Omnipath(
multinichenet_output_group1_t2vst1$prioritization_tables,
prioritized_tbl_oi_250_shared_E_NE)
plot_oi
```
### Interactions decreasing after therapy in the E group
Visualize now interactions that are in the top250 interactions for the contrast On-vs-Pre in the E group. We will focus on decreasing interactions here.
*E-group specific*
Of these, we will first zoom into the top10 interactions with biggest difference in the E-vs-NE group.
```{r}
group_oi = "G1.T1"
```
```{r}
prioritized_tbl_oi_250 = get_top_n_lr_pairs(
multinichenet_output_group1_t2vst1$prioritization_tables,
top_n = 250, rank_per_group = FALSE) %>%
filter(group == group_oi) %>%
inner_join(lr_network_all)
```
```{r}
ids_group1_vs_group2 = comparison_table %>% filter(group == "timepoint1" & id %in% prioritized_tbl_oi_250$id) %>% top_n(10, difference_group1_vs_group2) %>% filter(difference_group1_vs_group2 > 0) %>% pull(id)
```
```{r}
prioritized_tbl_oi_250_unique_E = prioritized_tbl_oi_250 %>%
filter(id %in% ids_group1_vs_group2)
```
```{r, fig.height=4, fig.width=24}
plot_oi = make_sample_lr_prod_activity_plots_Omnipath(
multinichenet_output_group1_t2vst1$prioritization_tables,
prioritized_tbl_oi_250_unique_E)
plot_oi
```
*shared between E-group and NE-group*
Now we will look at the interactions with a very similar score between E and NE. So these interactions are interactions that decrease after therapy, but shared between group E and NE.
```{r}
ids_E_shared_NE = comparison_table %>% filter(group == "timepoint1" & id %in% prioritized_tbl_oi_250$id) %>% mutate(deviation_from_0 = abs(0-difference_group1_vs_group2)) %>% top_n(10, -deviation_from_0) %>% arrange(deviation_from_0) %>% pull(id)
```
```{r}
prioritized_tbl_oi_250_shared_E_NE = prioritized_tbl_oi_250 %>%
filter(id %in% ids_E_shared_NE)
```
```{r, fig.height=4, fig.width=24}
plot_oi = make_sample_lr_prod_activity_plots_Omnipath(
multinichenet_output_group1_t2vst1$prioritization_tables,
prioritized_tbl_oi_250_shared_E_NE)
plot_oi
```
### Interactions increasing after therapy in the NE group
Visualize now interactions that are in the top250 interactions for the contrast On-vs-Pre in the E group.
To avoid making this vignette too long, we will not explicitly show this for the NE group since the code is the same as for the E group, you only need to change the `group_oi` to `"G2.T2"`.
### Conclusions of these comparisons
Whereas these comparisons are quite intuitive, they are also relatively arbitrary (requiring cutoffs to compare interactions) and suboptimal. They are suboptimal because the prioritization scores of interactions are relative versus the other interactions within a group. If there is a difference in __effect size__ of the therapy-induced changes in CCC, comparing the final prioritization scores may not be optimal.
Therefore, we will now discuss another way to infer group-specific therapy-associated CCC patterns: namely setting up complex multifactorial contrasts in the DE analysis of a MultiNicheNet analysis.
## 3B) E-vs-NE: MultiNicheNet analysis with complex contrast setting: (G1.T2-G1.T1)-(G2.T2-G2.T1)
*Set the required contrasts*
For this analysis, we want to compare how cell-cell communication changes On-vs-Pre anti-PD1 therapy are different between responder/expander patients vs non-responder/expander patients. In other words, we want to study how both patient groups react differently to the therapy.
To perform this comparison, we need to set the following contrasts:
```{r}
contrasts_oi = c("'(G1.T2-G1.T1)-(G2.T2-G2.T1)','(G2.T2-G2.T1)-(G1.T2-G1.T1)','(G1.T1-G1.T2)-(G2.T1-G2.T2)','(G2.T1-G2.T2)-(G1.T1-G1.T2)'")
contrast_tbl = tibble(contrast =
c("(G1.T2-G1.T1)-(G2.T2-G2.T1)",
"(G2.T2-G2.T1)-(G1.T2-G1.T1)",
"(G1.T1-G1.T2)-(G2.T1-G2.T2)",
"(G2.T1-G2.T2)-(G1.T1-G1.T2)"),
group = c("G1.T2","G2.T2","G1.T1","G2.T1"))
```
To understand this, let's take a look at the first contrasts of interest: `(G1.T2-G1.T1)-(G2.T2-G2.T1)`. As you can see, the first part of the expression: `(G1.T2-G1.T1)` will cover differences on-vs-pre therapy in group1, the second part `(G2.T2-G2.T1)` in the NE group. By adding the minus sign, we can compare these differences between the E and NE group.
### Run the analysis
In this case, the first 3 steps (cell-type filtering, gene filtering, and pseudobulk expression calculation) were run for the entire dataset and are exactly the same now as for the first MultiNicheNet analysis. Therefore, we will not redo these steps, and just start with step 4, the first unique step for this second analysis.
*Differential expression (DE) analysis: determine which genes are differentially expressed*
```{r}
DE_info = get_DE_info(
sce = sce,
sample_id = sample_id, group_id = group_id, celltype_id = celltype_id,
batches = batches, covariates = covariates,
contrasts_oi = contrasts_oi,
min_cells = min_cells,
expressed_df = frq_list$expressed_df)
```
Check DE output information in table with logFC and p-values for each gene-celltype-contrast:
```{r}
DE_info$celltype_de$de_output_tidy %>% head()
```
Evaluate the distributions of p-values: