The Practical Molecular Optimization (PMO) benchmark has long served as a standard for evaluating molecular optimization methods. However, its oracles are simple property-based functions that do not capture the complexities of structure-based drug design. As the field advances, there is a growing need for benchmarks that reflect real-world design challenges.
PMO-Dock, introduced in a recent arXiv preprint, directly addresses this limitation. The new benchmark integrates docking, specificity, and generalization into its evaluation framework, providing a more realistic and rigorous test for optimization algorithms. By shifting focus from abstract property scores to structure-aware metrics, PMO-Dock aims to better align benchmark performance with practical drug discovery outcomes.
The authors position PMO-Dock as a successor that builds on PMO's standardized evaluation while extending it to structure-based tasks. While the original PMO remains useful for quick property-based comparisons, PMO-Dock offers a more demanding and relevant assessment for methods intended for real-world design. The preprint does not yet detail experimental results, but the benchmark's design signals a clear move toward structure-aware evaluation in molecular optimization.