search

snap-stanford / GEARS

GEARS is a geometric deep learning model that predicts outcomes of novel multi-gene perturbations
MIT License
189 stars 38 forks source link

How to define pertubation? #43

Closed Larrycpan closed 7 months ago

Larrycpan commented 7 months ago

Thanks for your great work. I am writing to ask how to define the perturbation in GEARS study. Does it mean knockout or knockdown? For example, can GEARS predict the transcription profile after perturbing genes A and B simultaneously, where gene A expression decreases by 10% (e.g. 10 to 9) and gene B expression decreases by 20% (20 to 16)?

yhr91 commented 7 months ago

Thanks for your question. Under the current implementation, GEARS learns to model the perturbation type that is used in the training set. For example, if it is trained on knockout data, it will attempt to predict knockout effects. In the paper, we recommend that GEARS be trained on data generated under the same experimental conditions, including the same perturbation type.

It's possible to include additional variables during training to model different perturbation types such as activation versus repression, but that would need changes in the code as well as a training dataset that contains both perturbation types.

Larrycpan commented 7 months ago

Thanks for your question. Under the current implementation, GEARS learns to model the perturbation type that is used in the training set. For example, if it is trained on knockout data, it will attempt to predict knockout effects. In the paper, we recommend that GEARS be trained on data generated under the same experimental conditions, including the same perturbation type.

It's possible to include additional variables during training to model different perturbation types such as activation versus repression, but that would need changes in the code as well as a training dataset that contains both perturbation types.

Thanks for your reply!