Fig. 1
From: DeepSA: a deep-learning driven predictor of compound synthesis accessibility
An illustration of the designed architecture for DeepSA. We designed and trained Bidirectional Encoder Representations from Transformers (BERT) model and fine-tuned on our labeled synthesis related data to evaluate synthesizability from the given molecules’ SMILES. A dense layer was added following the BERT layer to perform binary classification task. If the given score is equal to or greater than 0.5, the molecule is considered difficult to synthesize; otherwise, if the score is less than 0.5, it will be easy to obtain