Open envest opened 1 year ago
One concern about using this method is training data leakage. Many (all?) of the data sets used to train medullo also appear in our study. We need to be intentional about limiting our medullo testing to data sets not appearing in their training set. This may take the form of a distinct test analysis on particular data sets that don't overlap with their study.
https://github.com/d3b-center/medullo-classifier-package can now work as a single sample predictor -- we need to include it as a comparison method and understand its functionality
Once Single Sample Classification PR (their # 6) is merged:
test_medullo()
inutils/modeling.R
withrun_one_model()
andrun_many_models()
Rathi, K. S. et al. A transcriptome-based classifier to determine molecular subtypes in medulloblastoma. PLoS Comput. Biol. 16, e1008263 (2020)