Perturbation simulation

[gilgolan73]

Hello,
I was going through the tutorial for TF perturbation (https://scenicplus.readthedocs.io/en/latest/Perturbation_simulation.html).
I wanted to ask- does the random forest regression model takes into account TFs that are negative regulators of the genes? or only positive regulators?
Because I see that the gene_to_TF object (given to the train_gene_expression_models function) includes both negative and positive regulators.

Thank you,
Gil

[SeppeDeWinter]

Hi @gilgolan73

Yes, it does take into account both positive and negative interactions.

Best,

Seppe

[gilgolan73]

Hi @SeppeDeWinter ,
Does it distinguish between them ? because the gene_to_TF object includes both of them without any information about whether they positively or negatively regulate the gene.

Thanks,
Gil

[SeppeDeWinter]

Hi @gilgolan73

Sorry I misunderstood your question.
No it treats both kind of TFs in the same way (i.e. this extra information is not explicitly provided as an extra feature).

Best,

S

[gilgolan73]

Hi,
OK. So do you believe it will be better to give the model only positive regulators ? (to get more accurate results for the perturbation simulation).

Thanks,
Gil

[SeppeDeWinter]

No, I think giving both is better.
The model will learn (again) that some have a negative effect. This might be important for predicting gene expression.

S

[gilgolan73]

OK, thank you for the answer