Fig. 1
From: A systematic review of deep learning chemical language models in recent era
Illustration of molecular representations and chemical language models. A displays various molecular representations of propa-2-one (acetone). B showcases RNNs as chemical language models and their autoregressive approach for generating chemical entities where [SOS] and [EOS] stands for start of sentence and end of sentence tokens, respectively. RNNs cells are also shown where the symbols •, *, and + denote dot-product, elementwise matrix multiplication, and addition, respectively. Each arrow corresponds to matrix multiplication utilizing a learnable matrix. Finally, in the context of RNNs, ‘x’, ‘a’, and ‘c’ correspond to the input, information matrix, and memory term, respectively. C illustrates the schematic representation of VAEs, while D presents the schematic representation of GANs. E displays the transformer model proposed by Vaswani and colleagues.31