ATC-SMILES

The benchmark ATC-SMILES is built for ATC classification. ATC-SMILES consists of 4545 compounds/drugs and their SMILES sequences. The benchmark is with the maximum coverage (81.34%) of KEGG dataset which contains all 5588 known drugs/compounds used for ATC analysis. Prior to this benchmark, the most widely adopted one is Chen-2012 which covers 3883 (69.49%) drugs in KEGG and is mainly used for generating inter-drug correlations (e.g. STITCH). The two benchmarks are compared in Table 1. ATC-SMILES is designed to be inclusive to Chen-2012, but there are 2.16% misalignment due to the mismatching of drug IDs that we will explain soon. ATC-SMILES can be extended with new drugs much easier than previous benchmarks as long as the SMILES sequences are available. Trails/experiments are not a must.