RascalMCES scores for benchmarking tasks (>5,000,000 pairs)

To compare the different similarity scores to a graph-based measure, we selected 5,557,963 unique pairs from the ms2structures dataset (see https://doi.org/10.5281/zenodo.15650044). Those were randomly drawn from candidates for all included compounds for which the mass difference was < 100 Da. RascalMCES scores were computed using RDKit on an i9-13900K CPU with the following settings: similarityThreshold = 0.05 maxBondMatchPairs = 1000 minFragSize = 3 timeout of 60 seconds. For 144,286 pairs, the computation timed out, and the resulting scores were not trusted. This resulted in a final benchmarking set of 5,413,677 unique compound pairs with computed rascalMCES scores.