Issues
- 0
- 0
- 0
determine number of really close ligands that were predicted my molrec, showing it can learn likely ligands without having access to structure
#37 opened by ljmartin - 0
Perhaps don't filter negatives from PubChem. They are all testing agonism (not antagonism)
#40 opened by ljmartin - 0
- 2
filter predicted ligands.
#38 opened by ljmartin - 1
download relevant assays from pubchem for molrec
#39 opened by ljmartin - 1
Get top predicted ligands in MolRec.
#19 opened by ljmartin - 1
try removing single-label ligands. Their latent vector should just be equivalent to the protein
#32 opened by ljmartin - 0
parse all interaction data from chembl
#31 opened by ljmartin - 1
Settle on best evaluation - including data, visualization, and recording/saving
#23 opened by ljmartin - 1
Fully evaluate label correlation, implicit, and lightFM on the time_split test
#24 opened by ljmartin - 1
- 2
calculate credible region around median
#29 opened by ljmartin - 1
save all ranks to raw
#30 opened by ljmartin - 1
- 1
Use masks to remove known ligands from rankdata
#27 opened by ljmartin - 1
put all HPOs into a single script
#26 opened by ljmartin - 1
shift training calls into utils
#25 opened by ljmartin - 1
Change HPO to use scipy.stats.mstats.rankdata, removing rows with no test data to save time.
#22 opened by ljmartin - 1
Inspection paradox
#15 opened by ljmartin - 1
- 1
- 1
- 1