search

antigenomics / tcr-pmhc-study

Mining TCR:pMHC structural data
5 stars 1 forks source link

EDA on the new loop data #10

Closed vadimnazarov closed 7 years ago

vadimnazarov commented 7 years ago

Split to test / train / CV + plot them

vadimnazarov commented 7 years ago

Split to equal number of real / putative CDRs

vadimnazarov commented 7 years ago

Split to equal number of "upside" and "downside" peptides for the training.

Check if "downsides" are "downsode" for both X and Y coordinates.

mikessh commented 7 years ago
  1. I prefer to treat MHC peptides separately, selecting peptide/antigen features that influence orientation wrt tcr-mhc axis
  2. It can pay off if we first select representative profiles using unsupervised clustering
  3. I would be really happy if we can outline loop/peptide length as a continuous parameter rather than fitting a model for every length