Analysis promising cell lines from enhancer test data

[selinaLa]

tasks:

• market_basket analyses
• differential analyses between all 3
• directionality analyses -> necessary in our case?? do not see use at the moment
• finding biological interesting TF co occurences (teilweise)
• find new cell lines and create new issue
  • (th1 ( Typ1 Helferzellen)/CD4+/CD8+)
  • small interstine/fetal small interstine *
  • PANC-1(cancer cell line of pancreas)/ pancreas*
  • hl60 (human leukemia cell line acute promyelocytic leukemia)/NB4 ( acute promyelocytic leukemia cell line)*
  • hep G2 (human liver cancer)/ liver *
• network solution implementation
• find out which treshold to use (for clearer network), or just adjusting each time

[selinaLa]

note: how to create network plot from differential object

[selinaLa]

Differential analysis of CD4+/CD8+, plotting heatmap
After function compare I want to display results in heatmap of CD4+ and CD8+. Then I get this result.

with the rules in the list

	TF1	TF2	CD4+_cosine	CD8+_cosine	CD4+/CD8+_cosine_log2fc	CD4+/CD8+_cosine_log2fc_pvalue	CD4+/CD8+_cosine_log2fc_pvalue_adj
AHR	ATF6	0.033026	0.139711	-1.304316	1.287173e-11	0.000002
ATF6	AHR	0.033026	0.139711	-1.304316	1.287173e-11	0.000002
RFX3	RXRB	0.029594	0.089801	-0.903453	7.935660e-07	0.127266
RXRB	RFX3	0.029594	0.089801	-0.903453	7.935660e-07	0.127266
NR3C1	TFEB	0.015293	0.058167	-0.833117	6.507370e-06	1.000000
...	...	...	...	...	...	...
FOS	FOSL2	0.361287	0.144311	1.124337	7.012899e-06	1.000000
JUN	JUNB	0.382696	0.153419	1.129647	2.713180e-06	0.435118
JUNB	JUN	0.382696	0.153419	1.129647	2.713180e-06	0.435118
JUN	FOS	0.376444	0.149213	1.139924	5.391554e-06	0.864654
FOS	JUN	0.376444	0.149213	1.139924	5.391554e-06	0.864654

When I use the function simplify the rules, so I dont get the duplicates, I get this list and this heatmap

TF1	TF2	CD4+_cosine	CD8+_cosine	CD4+/CD8+_cosine_log2fc	CD4+/CD8+_cosine_log2fc_pvalue	CD4+/CD8+_cosine_log2fc_pvalue_adj
AHR	ATF6	0.033026	0.139711	-1.304316	1.287173e-11	0.000002
RFX3	RXRB	0.029594	0.089801	-0.903453	7.935660e-07	0.127266
NR3C1	TFEB	0.015293	0.058167	-0.833117	6.507370e-06	1.000000
CTCFL	AHR	0.050248	0.119755	-0.826933	1.070661e-05	1.000000
RFX3	THRB	0.024469	0.073211	-0.816538	1.866323e-06	0.299306
...	...	...	...	...	...	...
JUN	FOSL2	0.374479	0.151570	1.115187	4.585157e-06	0.735331
FOS	JUNB	0.370528	0.148826	1.122216	6.176374e-06	0.990517
FOSL2	FOS	0.361287	0.144311	1.124337	7.012899e-06	1.000000
JUN	JUNB	0.382696	0.153419	1.129647	2.713180e-06	0.435118
JUN	FOS	0.376444	0.149213	1.139924	5.391554e-06	0.864654

Does it matter which I use? Because in both theoretically there is the same input. Only in the unsimplified heatmap the imfomation is mirrored and therefore doubled

[selinaLa]

Differential analysis of CD4+/CD8+, vulcano plot
in the documentation the selection line of the foldchange is a little over 1 or -1.

In my vulcano plot the selection line of the fold change is at 0,25 or -0,25.

Because of that I get a lot of rules and I get a pretty complex network. Is there a way to change the selection line border or is it generated of the rules?

[vheger]

• For the first question: Yes it contains the same information, only not mirrored. One may find the symmetrical plot easier to read, but the information content should stay the same. You could also check if the cluster remains the same (I can not check here, because the picture background is black and I can not see the clusters properly).

• For the second question: The thresholds are calculated automatically based on the distributions and pvalues (that's why the boundaries differ between the example and your data). For the pvalues a threshold of 0.05 is set as default, you can change this directly in the function (see attached picture). [y-axis]
  However, although the functionality to change the distribution percentage (as default also 0.05) used as threshold is given by the utility method (utils.get_threshold() ), there is no way to change the input as a user (DiffCombObj.select_rules() provides no parameter for this). [x-axis]

I need to check with Mette whether this is intended, otherwise we could add this.

[msbentsen]

Hi Selina, please have a look at the newest changes mentioned here: loosolab/TF-COMB#31 (comment).

So if you give measure_threshold_percent=0.01, you will get less rules and hopefully you are able to better visualize that in the network 👍