The original algorithm for predicting drug action phenotypes was embedded, which is based on the local active structure and can also provide the local active structure of serious adverse reactions for drug structure optimization. (See citation for details).
The original prediction algorithm was generalized to almost all compounds currently recorded in the PubChem database (about 110,000,000 in total), which covers 99.8% (168360/168756) of the currently discovered natural products.
The original prediction algorithm has excellent prediction performance. The statistical cross-validation AUCs were all above 0.8. In addition, the predicted results also had a high literature hit rate. (See citation for details)