First step of the A.I. for small molecule generation is to collect dataset that have SMILES of existing ligands. In order to train the model we should know proper protein structure to which small molecule binds. Then define its pocket 3D structure. This information is going to be sufficient to build 3D conditional inpainting GAN.
We expect to have results that include: 1. GAN model that can generate small molecule sequence. 2. 3D structure of the molecule with proper binding and positioning relative to protein pockets. 3. Eliminate binding with other human proteins to avoid unwanted side effects. 4. Can meet toxicology standards. 5. Generated molecule has high quality while testing in laboratory.