PyConDE 2018: "Designing better drugs with machine learning" by Daniel Kuhn

Optimizing small molecule drugs is a multiparameter problem

  • potency
  • robust synthesis
  • stability in vitro/vivo
  • safety
  • selectivity towards off-targets

Mission: design compounds by predictive models.

Challenge: which compounds to make next.

What changed in 10 years:

  • machine learning
  • more useful tools: pandas, RDKit, Jupiter notebooks.

The key is to combine different techniques and approaches.