GLOSS: A Multi-Strategy Sampling Algorithm for Optimization in Vast Chemical Spaces
- 1State Key Laboratory of Precision and Intelligent Chemistry, Hefei National Research Center for Physical Sciences at the Microscale, School of Chemistry and Materials Science, University of Science and Technology of China
- 2Hefei National Laboratory, Hefei National Research Center for Physical Sciences at the Microscale, University of Science and Technology of China
- 3Center for Scientific Intelligence Innovation
Abstract
Surrogate-based optimization accelerates closed-loop scientific discovery. In chemistry and materials science, however, sparse data over vast chemical spaces make surrogate predictions unreliable and undermine data efficiency. We develop GLOSS, a sampling algorithm that divides each batch across three complementary strategies: Global, Local, and Unexplored. Across multiple benchmarks, GLOSS reaches 95% of the optimum about twice as fast as Bayesian optimization and consistently outperforms genetic algorithms and random sampling, and its advantage grows more pronounced as the chemical space enlarges. By explicitly sampling unexplored regions rather than relying solely on surrogate predictions, GLOSS offers a scalable route to data-efficient closed-loop optimization in vast chemical spaces.
Related
- Chemical Editing of DNA-Mediated Nanoparticle Superlattices through a CRISPR-inspired Strategy
- A Novel Application of the Volumetric Absorptive Micro Sampling Strategy to Quantify Dexmedetomidine in Capillary Blood Using an LC-MS/MS Method.
- Optimization of Coagulation-Flocculation with Aluminum Sulfate for the Treatment of Chemical Industry Effluent
- QALPA: Property-guided diffusion modeling for efficient exploration of chemical spaces of flexible molecules
- Structure-Independent Genetic Algorithm-Based Molecular Optimization Using Pocket-Aware Molecular Representations
BibTeX
@misc{zhang2026gloss,
title = {{GLOSS: A Multi-Strategy Sampling Algorithm for Optimization in Vast Chemical Spaces}},
author = {Baicheng Zhang and Guoqing Zhang and Yi Luo and Jun Jiang and Zhuoying Zhu},
howpublished = {ChemRxiv},
year = {2026},
month = sep,
doi = {10.26434/chemrxiv.15008310/v1},
url = {https://chemrxiv.org/doi/full/10.26434/chemrxiv.15008310/v1},
note = {CC BY},
}
RIS
TY - UNPB
TI - GLOSS: A Multi-Strategy Sampling Algorithm for Optimization in Vast Chemical Spaces
AU - Baicheng Zhang
AU - Guoqing Zhang
AU - Yi Luo
AU - Jun Jiang
AU - Zhuoying Zhu
PY - 2026
DA - 2026/09/04
PB - ChemRxiv
DO - 10.26434/chemrxiv.15008310/v1
UR - https://chemrxiv.org/doi/full/10.26434/chemrxiv.15008310/v1
AB - Surrogate-based optimization accelerates closed-loop scientific discovery. In chemistry and materials science, however, sparse data over vast chemical spaces make surrogate predictions unreliable and undermine data efficiency. We develop GLOSS, a sampling algorithm that divides each batch across three complementary strategies: Global, Local, and Unexplored. Across multiple benchmarks, GLOSS reaches 95% of the optimum about twice as fast as Bayesian optimization and consistently outperforms genetic algorithms and random sampling, and its advantage grows more pronounced as the chemical space enlarges. By explicitly sampling unexplored regions rather than relying solely on surrogate predictions, GLOSS offers a scalable route to data-efficient closed-loop optimization in vast chemical spaces.
ER -