Fig. 6.

SMILES Data Augmentation results. The first row: We tested different values of n and report the gap between our evaluations schemes. ($\mu 1$ = random–one unseen, $\mu 2$ = random–both unseen, $\mu 3$ = one-unseen–both-unseen). Best n = 20 for Twosides and n = 80 for DrugBank. Second row: we compare ourselves, i.e. best-randomized smiles (Rand, smiles) to GIN, fingerprints, and canonical SMILES