The below map describes the organization of Makya modules inside a project.
In order to set up a fitting generator that helps them find promising drug candidates that fit the objectives of their project, users can customize their generators using the following tools:
-
Selecting rewards that will be used to guide generator optimization (using Reinforcement Learning); such rewards include:
- Similarity to a Dataset;
- QSAR Model predictions;
- 3D Shape and Pharmacophore Similarities to a reference ligand;
- Docking Score and Protein Interaction Fingerprint (Contact Score);
- External scores provided by third-party APIs.
-
Specifying chemical constraints (presence of some specific scaffold, presence or absence of certain moieties...);
- Through use of a Pains/Tox dictionary of forbidden motifs;
- Or through substructure constraints in the generation set-up.
- Setting up post-generation scorers to rescore molecules.
Generated molecules can then be visualized, analyzed, rescored, or exported for further analysis with external tools.