arXiv：cs.LG· Tomo Oga, Devesh Shah, Cailum M. K. Stienstra, Gabriel Asher, Antonio H. de O. Fonseca, Niall O'Connor, Michael Widrich·· 11 小时前
分子机器学习结构不相交划分随库规模漂移：MCES 聚类在 227,690 分子库上失效
Critical Mass: Structure-Disjoint Splits Drift With the Libraries They Are Built On
AI 导读
分子机器学习用结构不相交划分评估分布外性能，该规则基于 29,000 分子库设定，而公开库现已有 227,690 分子。作者提出优化版 MCES 准则，将 2.6×10^10 对分子缩减为 1.15×10^7 个整数规划、节省 2,259 倍，并证明保持无损划分。
正文
Abstract:Molecular machine learning evaluates out-of-distribution performance with structure disjoint splits, a construction rule fixed once at the scale its library happened to have: threshold a pairwise distance, then assign whole connected components to folds. For predicting structures from mass spectra, the field has converged on single-linkage clustering under the myopic Maximum Common Edge Subgraph (MCES) distance at a threshold of 10. This convention was set on 29,000 molecules but public libraries now hold 227,690 molecules. MCES is NP-hard per pair over a quadratically growing set, and so the criterion has never been evaluated at that scale. We propose an optimized version of this criterion, allowing for computationally feasible application on the public libraries: Admissible bounds and a union-find skip reduce $2.6 \times 10^{10}$ molecular pairs to only $1.15 \times 10^{7}$ integer programs, a $2{,}259\times$ saving, while provably producing the lossless partition. We find that at this scale the graph percolates. A single component holds $95.5\%$ of the library, capping any test set at $4.5\%$. What remains is not merely smaller, but systematically different: a held-out compound outweighs a training compound 88.1\% of the time, because a fixed absolute budget is a tighter relative tolerance on a larger molecule. Large library benchmarks using this splitting rule naively therefore conflate structural novelty with extrapolation to larger, more complex chemistry.
| Comments: | Accepted at 40th Conference on Neural Information Processing Systems (NeurIPS 2026). Workshop: AI Data Readiness for Scientific Discovery |
| Subjects: | Machine Learning (cs.LG); Computational Engineering, Finance, and Science (cs.CE) |
| MSC classes: | 68T05, 62H30, 92E10, 05C85 |
| ACM classes: | I.2.6; I.5.3; J.2; G.1.6 |
| Cite as: | arXiv:2610.12694 [cs.LG] |
| (or arXiv:2610.12694v1 [cs.LG] for this version) | |
| https://doi.org/10.48550/arXiv.2610.12694 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Michael Widrich [view email]
[v1]
Thu, 8 Oct 2026 20:20:41 UTC (1,063 KB)
来源：arXiv：cs.LG · arxiv.org