OptiMol: Optimization of Binding Affinities in Chemical Space for Drug Discovery

Ligand-based
drug design has recently benefited from the development
of deep generative models. These models enable extensive explorations
of the chemical space and provide a platform for molecular optimization.
However, the vast majority of current methods does not leverage the
structure of the binding target, which potentiates the binding of
small molecules and plays a key role in the interaction. We propose
an optimization pipeline that leverages complementary structure-based
and ligand-based methods. Instead of performing docking on a fixed
chemical library, we iteratively select promising compounds in the
full chemical space using a ligand-centered generative model. Molecular
docking is then used as an oracle to guide compound optimization.
This allows for iterative generation of compounds that fit the target
structure better and better, without prior knowledge about bioactives.
For this purpose, we introduce a new graph to Selfies Variational
Autoencoder (VAE) which benefits from an 18-fold faster decoding than
the graph to graph state of the art, while achieving a similar performance.
We then successfully optimize the generation of molecules toward high
docking scores, enabling a 10-fold enrichment of high-scoring compounds
found with a fixed computational cost.