Use logistic regression for identifying on- and off-morphology features via feature importance

[axiomcura]

This was a suggestion for an alternative method to determine which features should be classified as "on-" or "off-morphology" in a given cellular state comparison.

Since this is essentially a feature importance problem, machine learning models, particularly logistic regression, can be used to identify the most informative features. By treating the task as a binary classification problem (e.g., distinguishing between two cellular states), we can leverage the model’s learned coefficients to assess which features contribute most to the separation. Features with higher absolute coefficient values can be included in the "on-morphology" signature, while those with negligible or zero weights may be considered part of the "off-morphology" signature or dropped altogether.

In addition to this, users will be able to use normalized extracted features instead of feature selected profiles