Reaction-Dependent Predictability of Yields from Small Substrate-Scope Datasets
We evaluate structure-based regression using molecular fingerprints and ElasticNet across four reaction datasets (96–190 reactions). The models outperform constant-yield baselines for two transformations, including literature-assembled umpolung amide synthesis. Learning curves reveal reaction-dependent data requirements, with useful predictions emerging from 20 training examples in favorable cases.
Authors
- Takuma Morimoto (ORCID: https://orcid.org/0000-0001-9897-8338)
- Shinya Shiomi (ORCID: https://orcid.org/0000-0001-9906-8774)
- Masahiro Yoshida (ORCID: https://orcid.org/0000-0002-5595-0180)
Institutions
- Tokushima Bunri University (JP)
Publication Details
- Journal
- ChemRxiv
- Published
- 2026-11-17
- DOI
- https://doi.org/10.26434/chemrxiv.15000193/v2
- Primary Topic
- Computational Drug Discovery Methods
- Type
- preprint
Funders
- Ministry of Education, Culture, Sports, Science and Technology
- Japan Society for the Promotion of Science