Skip to main content
Fig. 2 | Journal of Cheminformatics

Fig. 2

From: COMA: efficient structure-constrained molecular generation using contractive and margin losses

Fig. 2

Overall performance on four benchmark tasks. a–d Evaluation of success rates over several similarity thresholds. For a given similarity threshold, the success rate is defined as the ratio of novel molecules satisfying both constraints property improvement and higher similarity than the threshold. e–f Comparison of the average scores of success rates over similarity thresholds. i–l Comparison of total performance scores. A total score is calculated as the sum of six metrics (validity, property, improvement, similarity, novelty, and diversity). a, e, and i, DRD2 results. b, f, and j, QED results. c, g, and k, Penalized LogP04 results. d, h, and l Penalized LogP06 results. Note that we did not evaluate two baseline models JTVAE and UGMMT on the two penalized logP datasets owing to the out-of-memory problems raised by the models during experiments

Back to article page