Optimization of a Graph‐to‐graph Translation Generative Model Under Multiple Constraints: Discovering New Melt‐cast Explosives

Developing new molecules whose desired properties must fit within multiple constraints is a longstanding challenge, particularly when experimental data is limited and costly. This multi‐objective optimization problem in the nearly infinite parameter space of chemical structures makes physics‐based design models difficult to develop. In this work, we apply a graph‐to‐graph translation model with property constraints to generate new molecules with an iterative approach. We demonstrate this methodology to suggest molecules for melt‐cast explosives that minimize volatility while also improving performance as estimated by oxygen balance and constraining on synthesizability. We identify training epochs and initial training set size as key factors in determining generated molecule diversity.