TotSyn: A Total Synthesis Reaction Dataset for Machine Learning in Organic Chemistry

Total synthesis is the construction of complex natural products from simple, commercially available starting materials. Despite its importance in synthetic chemistry, it remains underrepresented in machine learning resources, as existing reaction datasets are largely derived from patent data such as USPTO and fail to capture its distinctive chemical complexity. Here, we introduce TotSyn, a curated total synthesis reaction dataset collected from peer-reviewed journals. We analyze its reaction template diversity and longest linear sequence (LLS) and compare them with USPTO-50k, revealing substantial differences in template distribution and route complexity. Our results highlight the limitations of patent-derived datasets and position TotSyn as a benchmark resource for machine learning on total synthesis.