Fig. 3From: Enhancing chemical synthesis: a two-stage deep neural network for predicting feasible reaction conditionsAn illustration of negative sampling. The numbers in the blue and red squares represent the predicted probabilities assigned to each solvent and reagent label, respectively. Chemicals recorded in the reaction conditions in Reaxys are marked as positive labels (shown in orange). On the other hand, reagents and solvents not utilized in the actual reaction data but having predicted probabilities surpassing 0.1 are classified as hard negative labels (depicted in gray). This indicates that these chemicals are not actually pertinent to the reaction, yet the model might mistakenly consider them as feasible solvents or reagentsBack to article page