Finding molecules similar to mestranol

Rediscovering a known drug from a similarity score is the simplest check that a design algorithm can steer towards a target region of chemical space. Mestranol, a steroid, is the hardest of GuacaMol's three similarity targets, because its fused-ring scaffold is difficult to assemble by small edits.

Task

Propose molecules as SMILES strings and maximise similarity to mestranol, measured as Tanimoto similarity over atom-pair fingerprints and thresholded at 0.75, so anything above the threshold counts as full marks. Scored by AUC top-10 over a 10,000-call budget.

Background

The oracle is one of GuacaMol's similarity benchmarks, served by the Therapeutics Data Commons. In PMO's results it stays much harder than its siblings: albuterol similarity reaches 0.898 while nothing passes 0.63 here. The two are scored with different fingerprints, atom-pair for mestranol against FCFC4 for albuterol, and mestranol's fused steroid rings are harder to assemble by small edits, so the pair is a clean comparison of what current methods can and cannot reach.

Data and evaluation

• Data: there is no training set to fit. Any database a method needs is restricted to ZINC 250k, which ships with the repository, so no method gets an advantage from a richer corpus.
• Budget and seeds: 10,000 oracle calls per run, and five independent runs with different random seeds (python run.py METHOD --task production --n_runs 5 --oracles mestranol_similarity). Report the mean and standard deviation over the runs. The budget counts distinct valid canonical SMILES: a molecule queried twice is served from the cache for free, and an invalid SMILES scores 0.
• Metric: AUC top-10. The mean score of the ten best molecules found so far is recorded every 100 calls, the curve against the number of calls is integrated by the trapezoid rule and divided by the 10,000-call budget, so the value lands in $[0, 1]$ and higher is better. A run that stops early keeps its final top-10 value for the rest of the budget. Unlike a plain top-10 average, the metric rewards reaching a good score early.
• Official evaluation code: https://github.com/wenhao-gao/mol_opt (MIT), where main/optimizer.py holds the budget accounting and the AUC. Oracles come from the Therapeutics Data Commons via PyTDC (MIT); the ZINC database is free to use.

What is known

• Best in the PMO paper's table: GP BO at 0.627 ± 0.089, then REINVENT on SELFIES at 0.620 ± 0.029 and REINVENT at 0.618 ± 0.048. Six methods lie within 0.1 of the leader, so the ranking is not well separated.
• Later work on the same protocol reports higher: Mol GA at 0.762 ± 0.048 and Genetic GFN at 0.708 ± 0.057 (five runs, 10,000 calls).
• A random screen of ZINC 250k scores 0.409, so roughly two thirds of the leader's score comes for free from the database.
• For comparison, the same methods reach about 0.9 on the albuterol similarity oracle, which is what marks this one as the hard case.

What would count as progress

• A mean AUC top-10 above 0.627 on mestranol_similarity from five runs under the official protocol, reported with the per-run values and the standard deviation. To beat the field rather than the 2022 table, the number to pass is 0.762 (Mol GA).
• The run logs, which record every scored molecule in order, so the AUC and the oracle-call count can be recomputed rather than trusted.
• Code in the repository's method layout, so the result can be re-run on the other 22 oracles: a method that wins on one oracle and loses elsewhere is the pattern PMO was built to expose.

References

1. W. Gao, T. Fu, J. Sun and C. W. Coley, "Sample Efficiency Matters: A Benchmark for Practical Molecular Optimization", NeurIPS 2022 Datasets and Benchmarks (2022). https://arxiv.org/abs/2206.12411
2. N. Brown, M. Fiscato, M. H. S. Segler and A. C. Vaucher, "GuacaMol: Benchmarking Models for de Novo Molecular Design", Journal of Chemical Information and Modeling 59, 1096–1108 (2019). https://doi.org/10.1021/acs.jcim.8b00839
3. A. Tripp, G. N. C. Simm and J. M. Hernández-Lobato, "A Fresh Look at De Novo Molecular Design Benchmarks", NeurIPS 2021 AI for Science Workshop (2021). https://openreview.net/forum?id=gS3XMun4cl_
4. A. Tripp and J. M. Hernández-Lobato, "Genetic algorithms are strong baselines for molecule generation" (2023). https://arxiv.org/abs/2310.09267
5. H. Kim, M. Kim, S. Choi and J. Park, "Genetic-guided GFlowNets for Sample Efficient Molecular Optimization", NeurIPS 2024 (2024). https://arxiv.org/abs/2402.05961

Source

Imported from PMO (mol_opt), the mestranol_similarity oracle, curated by Wenhao Gao, Tianfan Fu, Jimeng Sun and Connor W. Coley. Open according to the source as of 2026-10-20. The statement above is written in our own words; see the source for the original wording, history and discussion.