Simulating a DMTA cycle

Code for simulating a DMTA cycle, with docking used as a proxy for the Test stage. The full workflow involves iteratively training an ML model, making predictions on a large dataset, selecting compounds for experimental/computational validation (in this case using docking to represent a binding affinity assay) and then retraining the models with the additional experimental/computational data.

• run_iterations_script.py: Main script to simulate the DMTA cycle. This can be run over multiple CPUs, with the model training, prediction calculation and docking steps all parallelised.

• run_iteration_fn.py: Functions required by the run_iterations_script.py. In particular this includes functions to ensure dataset files are only read/written to by one process at once.

• selection_fns.py: File containing different selection methods for choosing next round of compounds for docking and retraining ML models. Current methods include:

  • Highest predicted value
  • Highest prediction uncertainty
  • Diverse set of molecules

The repository also includes the directory structure used to store the results of the runs.