Can you fine-tune and perform perturbation analysis on the same dataset?
#228
by
Pingiotto
- opened
As you could probably tell by the question, I am no AI expert.
If I were to fine tune the model using a dataset, then use that fine-tuned model to perform in silico perturbation analysis on the same dataset, what would be the consequences (if any)? Wouldn't the model just be very good at its task?
Thanks for taking time to answer this silly question.
Thank you for your interest in Geneformer! The in silico perturbation can be considered a form of interpretability. For example, a fine-tuned model may perform well in classifying the given cell states. The in silico perturbation analysis will reveal the factors that drive the model's ability to distinguish the classes.
ctheodoris
changed discussion status to
closed