Meeting notes

24.09

• Small table without Seq2Seq

• Implement Seq2Seq

• Rerun with Seq2Seq

22.08

Report/any numbers?

31.07

• Task 3 - physical set
  • https://github.com/richlewis42/scikit-chem/blob/master/skchem/features/physicochemical.py
  • Run sth on server https://github.com/gmum/servers/settings
• Task 4 -
  • Upadte onlin and has a concatenation/report
  • Recommend: markdown document on github

run_grid.sh exp1: cmd1 (relu resnet on HIV) exp2: cmd2 (log reg on HIV) ... exp N: cmd N

./parallel run_grid.sh —n_jobs=10 GNU

Deepchem, smiles -> features

• Andril -
  • set up account

Task1: Template proposition:

• https://github.com/gmum/toolkit/tree/master/example_dl_project
• Refactor your current code into it

Task 2: Featurization task:

• 2 models (log reg, NN - smiles directly)
• 2 feature sets
  • maccs
  • molprint2d
• 3 datasets

Dataset:

• muv
  • tries to be unbiased
  • transfer learning doesnt seem to work
• hiv
  • ??
• tox21
  • transfer learning seems to work well

Later:

• Some more challenging

Related work:

• https://arxiv.org/pdf/1706.06619.pdf,