Overview
A screening pipeline can return many promising molecules that share the same chemotype. When only a small set can be tested, ranking by score alone can leave little chemical diversity in the final selection.
OPDiv formulates this selection as an integer optimization problem: choose a set of high-scoring molecules while enforcing a specified diversity threshold. The paper explores fingerprint, shape, and electrostatic distances and compares the resulting score–diversity tradeoffs.
The approach also provides a reference for evaluating screening and generative pipelines by the diverse compound sets they can produce, alongside their individual molecule rankings.