Closed incase-1 closed 1 month ago
Yes, for genes that were not present in the perturbation graph at the time of training, you would have to have to include them in the perturbation graph and re-train the model. Otherwise, the prediction will probably just be random.
Hi Team, I used the "Using Trained Model" code from your GitHub repository. When running "gears_model.predict([['VEGFR3']])", I encountered an error stating that "VEGFR3 is not in the perturbation graph. Please select from GEARS.pert_list!" As a result, I created a new gene list and passed it to the gene_set_path parameter. Do I need to retrain the model in this case?