Sitagliptin MPO: same properties, different molecule

Find molecules unlike sitagliptin that keep its formula, logP and polar surface area, in 10,000 calls.

shoal State the problem, from PMO (Practical Molecular Optimization) 3fa3b55 · 2周前历史

59 行 · 5.3 KB 原始文件永久链接历史

1	# Sitagliptin MPO: same properties, different molecule
2
3	A useful trick in medicinal chemistry is to keep a drug's physical profile while changing its structure, to escape patents or liabilities. This task formalises that: find molecules as dissimilar to sitagliptin as possible that still share its molecular formula, lipophilicity and polar surface area.
4
5	## Task
6
7	Propose molecules as SMILES strings and maximise the sitagliptin multi-property objective, the geometric mean of four terms: dissimilarity to sitagliptin (Tanimoto over ECFC4 fingerprints, scored by a Gaussian centred at 0 with width 0.1), $\log P$ near 2.0165, topological polar surface area near 77.04 and agreement with the formula $\mathrm{C}_{16}\mathrm{H}_{15}\mathrm{F}_6\mathrm{N}_5\mathrm{O}$. Scored by AUC top-10 over a 10,000-call budget.
8
9	## Background
10
11	The objective is one of the GuacaMol multi-property benchmarks of Brown and co-workers, here through the Therapeutics Data Commons; PMO notes that its implementation differs from GuacaMol's original, so numbers are comparable within PMO and not across the two. Because the four terms pull against each other, a geometric mean punishes any one failure, and this is the oracle where the PMO field collapses: the overall best method, REINVENT, scores 0.021 here while a simple genetic algorithm reaches 0.433.
12
13	## Data and evaluation
14
15	- Data: there is no training set to fit. Any database a method needs is restricted to ZINC 250k, which ships
16	with the repository, so no method gets an advantage from a richer corpus.
17	- Budget and seeds: 10,000 oracle calls per run, and five independent runs with different random seeds
18	(`python run.py METHOD --task production --n_runs 5 --oracles sitagliptin_mpo`). Report the mean and standard deviation
19	over the runs. The budget counts distinct valid canonical SMILES: a molecule queried twice is served from the
20	cache for free, and an invalid SMILES scores 0.
21	- Metric: AUC top-10. The mean score of the ten best molecules found so far is recorded every 100 calls, the
22	curve against the number of calls is integrated by the trapezoid rule and divided by the 10,000-call budget, so
23	the value lands in $[0, 1]$ and higher is better. A run that stops early keeps its final top-10 value for the
24	rest of the budget. Unlike a plain top-10 average, the metric rewards reaching a good score early.
25	- Official evaluation code: <https://github.com/wenhao-gao/mol_opt> (MIT), where `main/optimizer.py` holds the
26	budget accounting and the AUC. Oracles come from the Therapeutics Data Commons via PyTDC (MIT); the ZINC
27	database is free to use.
28
29	## What is known
30
31	- Best in the PMO paper's table: Graph GA at 0.433 ± 0.075, then STONED at 0.393 ± 0.083 and SMILES GA at 0.363 ± 0.057.
32	- REINVENT, the overall leader of the benchmark, scores 0.021 ± 0.003 here, and several strong methods are near zero, so this task is a counterexample to ranking methods by their average.
33	- Later work on the same protocol reports much higher: Mol GA at 0.677 ± 0.055 and Genetic GFN at 0.634 ± 0.039 (five runs, 10,000 calls), which means the task was far from its ceiling in 2022.
34	- A random screen of ZINC 250k scores 0.022, essentially nothing.
35
36	## What would count as progress
37
38	- A mean AUC top-10 above 0.433 on `sitagliptin_mpo` from five runs under the official protocol, reported
39	with the per-run values and the standard deviation. To beat the field rather than the 2022 table, the number to
40	pass is 0.677 (Mol GA).
41	- The run logs, which record every scored molecule in order, so the AUC and the oracle-call count can be
42	recomputed rather than trusted.
43	- Code in the repository's method layout, so the result can be re-run on the other 22 oracles: a method that wins
44	on one oracle and loses elsewhere is the pattern PMO was built to expose.
45
46	## References
47
48	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>
49	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>
50	3. J. H. Jensen, "A graph-based genetic algorithm and generative model/Monte Carlo tree search for the exploration of chemical space", Chemical Science 10, 3567–3572 (2019). <https://doi.org/10.1039/C8SC05372C>
51	4. A. Tripp and J. M. Hernández-Lobato, "Genetic algorithms are strong baselines for molecule generation" (2023). <https://arxiv.org/abs/2310.09267>
52	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>
53	6. K. Huang et al., "Artificial intelligence foundation for therapeutic science", Nature Chemical Biology 18, 1033–1036 (2022); the Therapeutics Data Commons, which supplies the oracle implementations. <https://doi.org/10.1038/s41589-022-01131-2>
54
55	## Source
56
57	Imported from [PMO (mol_opt), the `sitagliptin_mpo` oracle](https://github.com/wenhao-gao/mol_opt), curated by Wenhao Gao, Tianfan Fu, Jimeng Sun
58	and Connor W. Coley. Open according to the source as of 2026-10-20. The statement above is written in our own
59	words; see the source for the original wording, history and discussion.