Figure 4.

The cosine similarity scores and t-SNE projection of different feature representations among 30 plant-derived compounds based on (a) chemical SMILES embedding vectors, (b) GO functional embedding vectors, and (c) global embedding vectors. Compounds with a higher probability score predicted by our model, and also, direct evidences in CTD and published reports were shown in red circles. This with a higher probability score but without evidence supported were in blue circles, while the rest with a low probability score were represents by green triangles. CTD: Comparative Toxicogenomic Database; GO: Gene Ontology; SMILES: Simplified Molecular-Input Line-Entry System; t-SNE, t-distributed stochastic neighbor embedding.